Week of 2/1/99

Search TTG - Robert Morgan pointed out last week that he wasn't even aware that a search page existed. The only place it appeared was on the Triad Technology group home page. So, I've added a link to the Search page. Note, however, that I've been having problems lately with the Search function. The web server that my site is running on is so busy that sometimes the Search simply times out with a "web server too busy" error message.

Special Reports - I'll be doing more reports in the future. These will be too large to embed in my Daynotes page, so I've created a Special Reports Home page and added links to it. At first, I'll just use this page to consolidate older reports that already exist scattered throughout the site. As things progress, I'll rearrange this page into a hierarchical home page, with links to product evaluations, system build logs, how-to reports, and so on.

I did a lot of reorganizing of the site yesterday, although much of it isn't yet visible to readers. But in moving stuff around, creating a new folder structure, etc., I ended up "changing" many older files, at least from Frontpage's point of view. The upshot is that many older files now have yesterday's date on them. If you load them, they'll appear to refresh as though they contained new material, but they don't. Only the regular stuff--today's Daynotes, the index page, etc.--really has new stuff. The rest of it may have a couple of new links (to the Search page, for example) or old links that point to files now located in new folders. So, don't trust your refresh today to tell you what's new and what isn't.

* * * * *

Jerry Pournelle had this to say on his web site last Saturday about the time he spends working on the site:

"Roberta says I spend too much time here, and that I start in the morning, which means that everyone else gets me thinking about what THEY want me to think about rather than what I want to think about. She has a point, so I think I'll start doing this work at night, after a day's work."

Roberta Pournelle is a very smart woman. I have the same problem here, although not to the extent that Jerry does. There are many days that I burn two or three hours first thing in the morning working on the web site, and I can't afford to spend that much time on it. In the past, I've been updating my site regularly in the morning, with occasional updates during the day and evening. Starting this week, that's going to change. I'll do my primary update in late afternoon or during the evening, say 17:00 to 23:00 my time. I'm Eastern Time (GMT -0500), so the update will typically occur between 22:00 GMT and 04:00 GMT.

I'd actually tried going this route a couple of months ago. What I ran into was server response problems later in the day. My web hosting company is located in California (GMT -0800). If I updated my web site by about 9:00 or 10:00 a.m. my time (6:00 or 7:00 a.m. theirs), the update proceeded normally. If I waited until after open of business on the west coast, the update would frequently blow up. I assumed that this was because the local load on their servers increased once the business day started out there, but who knows? Most of the timeouts occurred after the updates were already published to the web server and while FrontPage was doing maintenance and cleanup chores. Such timeouts don't appear to damage anything, so I may just live with them.

At any rate, I'll try doing updates in the afternoon, starting this afternoon. If the timeouts become too bad, I'll shift to doing them in late evening, once the people in California have quit work for the day. If that doesn't work, I may actually write the material the preceding evening and wait until the following morning to publish it.

* * * * *

This from Peter N. Linden [mailto:lindprof@ix.netcom.com] regarding controlling children's use of the phone:

We would like a security system which requires a security code before any outside call be placed. Do you know of anything like this. This would probably be a hardware/keypad electronic device.

Sure, but they're not consumer items as far as I know. And that's the problem. I know any number of telephone supply distributors that sell devices to block or restrict outbound long-distance calls on a per-line basis, but they don't sell to individuals.

The units I've seen are made by Viking and similar companies, and hook directly into the CO line at the demarc. The premise wiring then connects to the device. Most of them are single-line devices, so if you need to protect multiple lines, you'd need one for each line. Some of the very inexpensive devices simply block any call that begins with a "1" or a "0", which means they also block toll-free calls. Better ones are more configurable (for example, they may allow you to block or allow specific area codes), and allow you to override the block by entering a code on the telephone pad.

If you want to get a device like this, the best bet would be to call a local company that specializes in business telephone systems and tell them what you need. They'll probably sell it to you, but they may want to install it. I'd recommend that you let them do it if the price is reasonable. It shouldn't take an experienced installer any more than half an hour to hook it up at the demarc.

Before you buy something, check to see if your local telco offers such a service. I'll also post this in case one of my other readers knows of a consumer source for such devices.

* * * * *

This from Robert Morgan [mailto:robert@morgan.pair.com] regarding pair.com mail policies (the final paragraph actually came as a separate mail message, but I combined them for brevity):

Today [Sunday, 1/31/99 RBT] you mentioned your mail problems again, and also pair's mail handling policy. Pair forwards any and all mail addressed to your domain to one pop account. I use this feature when giving my mail address to any site: I just use a new address on the fly so I can find out if the site is giving away my mail address (ie. if I didn't know your site and needed to register I'd probably use ttgnet@morgan.pair.com).

Then, I set up filters on my mail program (still Eudora pro) to place mail addressed to each address in its own folder.

Also, if you choose pair's basic account or better, you can write a .forward file in your home directory on the pair server to arbitrarily forward any mail to anywhere else, including a file, or /dev/null.

A good solution is a linux box running the fetchmail program. That will get all the mail from pair's single pop mailbox and place the mail in individual pop mailboxes on the linux box. Presumably it would run round the clock, and your workstations would send and receive mail through the linux machine.

If you ultimately decide to go for a true SMTP mail host on one of your NT boxes, I can recommend Qualcomm's Worldmail server over MS Exchange. Exchange is a real dog for small offices. Worldmail is inexpensive ($159 for ten users).

A quick net search also came up with Cmail from www.computalynx.com, which is a mail server running on Win95/NT, and will fetch mail addressed to multiple addresses from one pop account. More expensive - 149 pounds.

I'm not completely sure that pair.com mail policies will make them a non-starter for me. Basically, the way I'm setup now at BigBiz is with one actual POP mailbox, whose name happens to be ttgnet@ttgnet.com. I have many mailboxes (thompson@ttgnet.com, robert@ttgnet.com, rbt@ttgnet.com, webmaster@ttgnet.com, info@ttgnet.com, etc.) that autoforward to my single mailbox.

I also have several mailboxes (barbara@ttgnet.com, steve@ttgnet.com, craig@ttgnet.com, etc.) that forward to external mailboxes. For example, barbara@ttgnet.com forwards to our dialup account at bellsouth.net, from which she POPs. Steve's and Craig's forward to their own local ISP accounts. I also have my mail configured so that any mail to a non-existent *.ttgnet.com account (e.g. moron@ttgnet.com, which I actually got one addressed to) is forwarded to my main mailbox.

If I understand you correctly, pair.com mail policies would allow me to do the same thing without paying for additional mailboxes, right? If that's true, my major objection to pair.com disappears. Of course, my main objection to BigBiz has already disappeared. Their connectivity is now normal, although the server they're running me on is still slow. I just used the TopHosts.com hosts connection testing page to check out the responsiveness of pair.com and www.ttgnet.com. They scored 99/100, and I got a 98/100.

As far as installing Linux and fetchmail, you're correct. That'd be the way to do it if I wanted a local mailserver. I don't think I'd buy an SMTP smart mailer for NT, though. I'd just install sendmail on Linux, although I might also install the Microsoft SMTP Server (a mail relay) on one of my NT boxes.

* * * * *

Afternoon: In re-reading my reply to Peter N. Linden about long-distance call blocking, it struck me that I wasn't clear when talking about the local telcos. When I suggested that that the local telco might provide such a service, I wasn't referring to them installing call-blocking hardware. What I meant was that they might be able to enable that function in their own switch for Mr. Linden's line(s), and simply charge him a monthly service fee for doing so. Some telcos have this service available and others don't. All or nearly all of them will block long-distance calls, but the override function may or may not be available.

* * * * *

And I note that FrontPage isn't doing as good a job as I thought in converting readers' email addresses to mailto: links. When I clicked on the link to Mr. Linden to send a clarification, I noticed that it had butchered the address by adding a second mailto: I've gone back and fixed the mailto: links here, but there may be some I missed elsewhere. My readers are smart enough to figure them out, though.

* * * * *

This from Bo Leuf [bleuf@algonet.se]:

BTW, thanks for posting those links to the short-list webhosts. I found both and the comparison I made between the two most interesting and enlightening. It clarified a somewhat vague picture I had developed about how webhosting is structured these days, and confirmed the notion that most of the "hosts" one sees are in fact space resellers, one or more rungs down in the hosting hierarchy.

The number of actual physical hosts, server farms, may turn out to be much lower than one imagines.

There are certainly many web hosting services that simply resell space. In fact, I could do that with my own account if I wished. But there are certainly hundreds of services that run their own servers, although many of them are co-located in glass houses that belong to other companies. For me, the sine qua non of a web-hosting company is its connectivity. Obviously, that's necessary but not sufficient. But without great connectivity, the best prices, policies, and support are worthless.

* * * * *

And more from Robert Morgan ]robert@morgan.pair.com] about pair.com and mail:

I wasn't aware of the Tophosts site, interesting. In their top 25 list, they rank pair.com second. Not bad at all.

You do understand me correctly. A .forward file in your home directory can selectively forward mail based on the header. I have the ftp account, and I just tried it and it worked. Well, I didn't test selective forwarding, I forwarded all my mail to another box. This is all documented in sendmail. I've forgotten much of it.

See http://pair.com/pair/chart.html, Note 5 at the bottom, which confirms free forwarding and autoresponders.

The nice thing about this business is that you don't need to remember the how's, just what's possible. See http://support.pair.com/howto/mailfwd.html to read the specifics on mail forwarding.

Yes, it appears that I may be able to get along with the pair.com mail policies. Just to confirm, I sent them another query letter this morning to ask specificially how they charge for mailboxes. I don't mind, for example, paying a buck a month to get Barbara her own POP mailbox, or even paying two bucks a month to get mailboxes for Barbara and me. What I wanted to make sure about was that I wouldn't end up having to pay fifteen or twenty bucks extra a month to cover all the other mailboxes I use for various purposes. It looks like I won't have to do that.

* * * * *

Friday at 12:18 p.m. I sent pair.com a query message. I got an autoresponder message in reply immediately that said that they take care of existing customers first and that their response to queries about new service might be delayed for a few days. I got the following response from pair.com tech support at 11:18 a.m. this morning, which I think is pretty fast service:

1. If I sign up for Advanced hosting, I assume that you would provide one static IP address, and the three hostnames www.ttgnet.com, mail.ttgnet.com, and ftp.ttgnet.com. Is this correct?

You are correct.

2. The cost of hosting my domain using your Advanced service appears to be $15 Setup, $16.95/month for the service, $25 setup for the virtual domain, and $1/month for the virtual domain. I'm unclear about what you waive for transferring an existing domain. Do you waive the $15 account setup, the $25 domain setup, or both?

We do not waive either of the above fee's. We do not charge extra for generating the transfer paperwork or processing the transfer on our end.

3. I write computer books for a living, and would plan to POP my mail from your server. Occasionally, people send me rather large files (e.g. a chapter manuscript zipped with figures). At various places on your site, you mention limits on mailbox size of 500 KB, 1 MB, and 3 MB. Which, if any, of these are hard limits? I normally POP my mail every few minutes during the day, but there is a chance that mail in excess of the above sizes would accumulate overnight. This would happen relatively infrequently, but it would be nice if the limits were soft.

We do not enforce the above storage limits unless things become a problem (seldom). The only hard limit is on incoming mail. Mail in excess of 2.5Mb will not always be accepted by the mail server.

4. If I choose to host (as opposed to park) another domain (e.g. my hardwareguys.com domain) on this same account, am I correct in assuming that: (a) that second domain is assigned its own static IP address, (b) that you provide the three hostnames www.hardwareguys.com, mail.hardwareguys.com, and ftp.hardwareguys.com, (c) that the cost to do this would be $25 setup and $1/month for service, and (d) that both domains would share the available 40 MB of disk storage and the 200 MB/day throughput?

You are correct.

5. If I sign up for your service, I want to minimize interruptions to my existing website readers. Can you provide me with my new IP address initially, which I can them provide to my readers on my current site so they can at least get to my new location by entering the IP address? I would then delay changing my InterNIC record for a few days. I doubt that BigBiz.com will be very helpful about changing the TTL on their nameservers to help the change proliferate quickly. Or is there something else you can suggest that will minimize the service disruption when I move to your service?

The best way to do this is to setup the account, domain name, and web pages on our servers first. Once that is done, send the transfer template that we generate for you to InterNIC. As long as you send the template from the e-mail address "thompsrb@BELLSOUTH.NET" the transfer should not involve your old host. InterNIC just needs to see that the change came from either the Administrative or Technical contact.

If you have any other questions, let me know.

So I sent them a followup query just to make sure I understood their mail policies.

* * * * *

This from Tom Syroid Tom Syroid [tsyroid@home.com] :

Read your notes on web updates with interest today..

As you know I've had my share of problems over the past 2 weeks, and while the solutions are not clearly focused yet, I suspect a good deal of my grief is coming from the time-of-day I update. There are other factors in the puzzle, of course, but those pieces are the hazy part still floating around in my head.

My web servers -- and the net in general of late -- have been incredibly slothful between 7pm and midnight (Saskatchewan local), which leads to spotty FTP uploads; always slow and sometimes incomplete due to time-outs. Which in turn leads to incomplete page updates.

Try your schedule out, but I suspect you will encounter some of the same troubles I have and end up going to your Option C: Write your page updates one day, post them early in the morning of the next. This is the route I'm going this week -- we'll see what comes of it.

Yep. The problem with figuring out what's causing web server delays is that there are so many variables. The connection the client has to his ISP; the connection from that ISP to the backbone; the connection from the web hosting provider to the backbone; general backbone congestion; the load on the web server itself, etc. etc. Broadly speaking, you can break down the problem into being caused by either connectivity, web server performance, or both.

In my case, until recently I had no problems with BigBiz on either score. Except that my problems updating my site during business hours were probably being caused by inadequate response time on the server itself. The mess late last week resulted from both problems. BigBiz was having connectivity and router problems *and* their web server was running very slowly. At this point, the connectivity problems appear to be resolved, but I'm still running on server #1, which is apparently overloaded. BigBiz offered to move me to server #5, which is much more lightly loaded. If I stick with them, I'll accept that offer. But I may well be moving to pair.com. If that's the case, I don't want to put BigBiz through the effort of relocating me to server #5 for such a short time.

Late Afternoon: My mother had a dentist appointment at 1:30 to have some teeth extracted to prepare for her getting dentures. Barbara took her over and brought her back about 3:00 p.m. She seems to be doing well. Barbara and I had spent the morning working on the ATX test bed system, but hadn't gotten it finished. When they returned and we got my mother settled in, Barbara and I went back to working on the system. We got everything installed and connected, did one final quick check to make sure we hadn't left a screwdriver in the patient, and then did the smoke test. Nothing happened. Nothing. The fan didn't even spin. We verified all the connections, and tried again. Still nothing. At this point, I'm not sure if the power supply is bad, or if there's a problem with the system board. I have some EPoX system boards on the way, which I hope will arrive this afternoon. I'll try one of them before I pull the power supply and take it back to CSO for a replacement.

I'm going to try posting this update to the server now (about 16:30 EST). We'll see what happens. If it posts, fine. Otherwise, I'll try it after dinner, and then again before bed if necessary. If none of those work, I'll post it first thing in the morning. Whatever happens, this will be the last update until tomorrow (Tuesday) afternoon or evening.

16:30 EST - I tried and failed to update the web site. I started the publication process, which seemed to begin normally. After about a minute, Front Page returned a server error: "Web <root web> is busy. Please try again." So I tried again, with the same lack of results. I'll give it a try after dinner.

17:15 EST - Decided to try to update the web site again before dinner. I started the publication process, which seemed to begin normally. Ordinarily, the publication process takes perhaps two minutes, and the modem lights flash pretty constantly during the early phases. This time, the modem lights would flash quickly for a second a two, followed by a minute or more of no activity.After ten or twelve minutes, the process finally timed out with the server error described above.

18:00 EST - I'm going to try again... Well, that one worked, but it took twenty minutes--from 18:02 to 18:22--to get the job done.

Tuesday

Tuesday, February 2, 1999

UPS showed up at 6:00 p.m. yesterday with a box from EPoX, a first-rate motherboard manufacturer. Although they're perhaps not as well known as some of their competitors, I've had nothing but good experiences with EPoX motherboards, and recommend them frequently. I've never heard anyone have anything bad to say about EPoX, and many people swear by them. One measure of my confidence in EPoX is that I bought an EPoX EP-68LXR, a Baby AT Pentium II board with an Intel 440LX chipset, to build Barbara's main workstation around.

So when I started to look around for motherboards to evaluate as possible candidates to recommend in my books, I naturally contacted EPoX. They sent me three motherboards, an EP-BXT (Slot 1 BX ATX board with integrated i740 video and Yamaha sound), a KP6-BS (dual Slot 1 BX ATX board), and an EP-58MVP3C-M (Socket 7 MVP3 100 MHz AT board). I thought about using the EP-BXT to replace the Intel Seattle-1 SE440BX in the test bed unit to verify that I indeed have a bad power supply, but I was pretty tired. That can wait until tomorrow. You'll be hearing more about these boards in the Special Reports section soon.

* * * * *

This from Joline Newman [jnewman@inav.net] regarding how to make a bomb:

I am very intrested in pyrotechnicians, and bomb making could you please e-mail me with some information and proceeders on how to make some hame made explosives, I would be most abliged.

My best advice is, don't.

When I played around with bombs and pyrotechnics as a teenager 30 years and more ago, things were very different. Nowadays, even if you somehow avoid blowing yourself up, you're likely to end up in jail. I saw a report recently of a man who was arrested and jailed, with bail set at a million dollars, simply for manufacturing cherry bombs and selling them to his co-workers. In the aftermath of Oklahoma City, the ATF and other law enforcement agencies have even less toleration for people playing around with explosives than they used to have, and that wasn't much.

Do yourself a big favor and don't even think about trying to making things that go boom.

* * * * *

For some reason, tonight seems to be a popular time to ask me how to do illegal things. This from Murilo Maia [mmaia@e-net.com.br]:

Who much make one silencer for Taurus PT-58 .380 ACP????

You can help me???

Eu sou pesaroso que eu não posso responder a esta pergunta. Alguns fabricantes unidos dos estados fazem silencers da pistola, mas seus venda, uso, e possessão são restringidos firmemente nos estados unidos. Se os silencers fossem legais em Brasil, sua loja local dos firearms seria provavelmente uma fonte boa. Se os silencers não forem legais em Brasil, eu recomendo que você os evita.

(I am sorry that I can't answer this question. Some United States manufacturers make pistol silencers, but their sale, use, and possession is tightly restricted in the United States. If silencers are legal in Brasil, your local firearms store would probably be a good source. If silencers are not legal in Brasil, I recommend that you avoid them.)

* * * * *

The following from Mitch Armistead [mhaathome@worldnet.att.net]:

Love your site. Any thoughts, or comments on AGP ports with Windows NT. Everything I found at Microsoft says that it's not supported till NT 5.0, though that seems not to be the case in Reality. Anything will be appreciated.

Thanks. As far as AGP support in W2K, I don't know. I have had various builds of Windows NT 5 and Windows 2000 installed here, although I don't have it running on anything at the moment. All of the machines I've installed it on have had PCI video, so I haven't had the opportunity to test AGP support in the betas. I have heard, although I don't know whether it is true, that Microsoft already has both AGP and USB support implemented for Windows NT 4, but decided not to ship it in Service Pack 4. The theory goes that they decided to withhold it as an added incentive for people to upgrade to W2K. My best guess is that we'll see full AGP support implemented in W2K.

* * * * *

The following from Warrick M. Locke [warlocke@mail.wf.net]:

Don't know how relevant this is, but the Web has been extremely slow for me over the last two weeks. Connect times averaging 3 minutes, sometimes no connect at all for 10 minutes at a time.

I'd thought it was my ISP (wf.net, MIL-TEL Communications, Wichita Falls, TX, talking over Southwestern Slow^h^h^h^hBell's voice network) coupled with my archaic computer. (I'm just a browser; I've nothing to put on a web site.)

Is there some new net-virus going around? Or is MCI having hiccups?

I don't know, but the web indeed seems to be running pretty slowly lately. We're just fortunate that packet switching in general and TCP/IP in particular are as well thought out as they are. They degrade gracefully under load. If they didn't, we'd all be getting the IP equivalent of constant busy signals.

* * * * *

Late afternoon: Another frustrating day. Barbara and I started the day by replacing the Intel Seattle-1 SE440BX motherboard in the ATX test-bed system with the EPoX EP-BXT motherboard that just arrived yesterday evening. We got everything connected, powered it up, and still nothing worked, not even the fan. I called Computer & Software Outlook, and they told me to bring back the power supply and they'd pull a new one from stock.

Then my mom's cleaning lady called up the stairs to tell me that she'd flushed the toilet and it'd backed up to nearly overflowing. That happens periodically when the switch in the ejector pump gets clogged. Ordinarily, I just jiggle one of the pipes to the pump and it starts running. I went down and jiggled the pipe with no results. Jiggled harder. Still no results. I figured that maybe a breaker had blown, so I went to check the breaker panel. No breakers had blown.

So I went into the finished area of the basement and unplugged a working lamp. It was lit when I pulled the plug. I took that back to the ejector pump and plugged it in. No light. Flipped switch a couple of times. Still no light. Aha. We have a bunch of GFCI receptacles. Since no breakers had blown, I figured it must be one of the GFCIs. I went around and checked and reset all of them. No joy. I took the lamp back into the finished area of the basement and put it back where it had come from. I didn't light there, either. Flipped the switch a couple of times. Still no light. Okay, everthing has to die sometime, and strange coincidences while troubleshooting are nothing new.

I grabbed the DirtDevil hand vacuum and plugged it in the the same receptacle as the ejector pump. It ran fine. That's depressing. A dead receptacle is easy to deal with, but now it looks like I need to call the plumber. I called him, and he showed up an hour or so later. I got them started and headed back upstairs. I have WinGate set to redial 99 times and when I sat down back at my desk it was sitting there redialing. Okay, that made sense. My modem doesn't connect directly to a phone line. Instead, it connect through my telephone system. I looked over at my desk phone and, sure enough, it was dead.

I figured that the plumbers had thrown the breaker before working on the ejector pump, and it just happened to be the same circuit that my telephone system is on. I tried to kill the WinGate dialer, but it refused to dial. Finally, after listening to it try to dial every five seconds for about 15 minutes--surely much more than 99 times--I headed downstairs to ask the plumbers if they'd mind just unplugging the pump and turning its breaker back on. They hadn't thrown the breaker to start with. Something had caused a GFCI to blow, and it happened to be the one that my telephone switch is connected to. I reset the GFCI, and both phones and modem started working again.

And when the modem started working, mail started arriving again. To add to the day's weirdness, here's what showed up (name and address omitted), along with my reply:

I just rec'd my M/C bill and lo and behold I find a $95.00 charge from you folks dated 12/25/98. My name is [name removed by RBT] and I am requesting you to please review and advise me of what I (didn't) purchased. Thank you, [address removed by RBT].

Huh? We don't even have a merchant account (and never have had one), so we can't take credit cards. You must be thinking of someone else.

The good news is that the plumbers suggested a fix for our occasional backflow problem. They headed out to pick up a 2" check valve to stop the backflow, and about five mintues later Barbara got back home from her errands. She has a new power supply, which I sincerely hope works. A few minutes after she got home, she took our four year old Border Collie Duncan on his long walk. A few minutes after that, she came running back in the house to shout that Duncan had been attacked and was injured.

As they were walking, Duncan stopped to stick his snout down a storm drain, a habit of his that we can't break. Barbara told him to come and kept walking. Duncan disobeyed and stayed at the storm drain. Barbara turned around just in time to see something come up out of the storm drain and go for Duncan's face. She described it as "a monster" and said that it snarled ferociously. She said it was brown, bigger than a squirrel, and smaller than a leopard, but couldn't otherwise describe it. In retrospect, she thought perhaps it was a raccoon or a a very large cat. It went for Duncan, he went for it, and when she finally got Duncan pulled back, his snout was "dripping blood". Barbara was terrified, thinking that Duncan had been seriously injured.

As it turned out, he wasn't injured, but you should see the other guy. By the time they got home, no blood was evident to me. Apparently, all the blood originated with the monster, and Duncan must have licked it off his snout. I called the vet and made an appointment this afternoon for them to look at Duncan (who appears to have no injuries) and give him booster shots for whatever. Barbara was afraid that whatever was down there was still alive and might attack a child or another pet, so she called animal control to report a vicious animal. Personally, I think it was probably a cat that made the mistake of fighting instead of running. Any cat that stands and fights a dog deserves what it gets. The jaws of a 70 pound dog are nothing to be taken lightly.

This is one of those days when I just won't be able to get much done.

Since I can't get any writing done for all the interruptions, I decided to swap out the power supply. Four screws, and two minutes later, the system boots. The moral here is, don't use cheap power supplies. I wouldn't have wasted a couple of hours and Barbara's trip back to CSO if I'd started with a good power supply in the first place. Unfortunately, good power supplies aren't cheap items. The PC Power & Cooling 300 watt ATX unit I'd like to put in this machine costs $120. Not many people will spend that on a power supply, but in my experience it's a worthwhile investment.

Later Afternoon: It's 16:30 now, and I'm whacked. The plumbers are back, and Barbara is getting ready to leave to take Duncan to the vet. The plumbers had to go four places to find a replacement switch for the ejector pump. I'm not surprised it needed replaced. It must be seven years old by now, and it's spent its life submerged in sewage.

I'd like to spend the evening reading, but my unread stack is almost non-existent. I'll watch Buffy tonight (it's a re-run, but they're all new to us) and then vegetate with a junk novel.

* * * * *

And I got the following response to my query from pair.com tech support:

Thanks for the quick response. I do have one final question, this one about mailboxes. Basically, the way I'm setup now at BigBiz is with one actual POP mailbox, whose name happens to be ttgnet@ttgnet.com. I can freely create mailboxes within my own domain (thompson@ttgnet.com, robert@ttgnet.com, rbt@ttgnet.com, webmaster@ttgnet.com, info@ttgnet.com, etc.) that autoforward to that single mailbox, from which I POP. I also have several mailboxes (barbara@ttgnet.com, steve@ttgnet.com, craig@ttgnet.com, etc.) that forward to external mailboxes. For example, barbara@ttgnet.com forwards to our dialup account at bellsouth.net, from which she POPs. Steve's and Craig's forward to their own external local ISP accounts. I also have my mail configured so that any mail to a non-existent *.ttgnet.com account (e.g. moron@ttgnet.com, which I actually got one addressed to) is forwarded to my main mailbox. If I understand you correctly, pair.com mail policies would allow me to do the same thing without paying for additional mailboxes, right?

Once you have a domain name setup for the account, all mail directed to the domain will be sent to your main mailbox. You can then forward specific messages as you choose. No cost.

What I don't know is whether with pair.com I can POP directly from "virtual" mailboxes that I create in the ttgnet.com domain. With BigBiz, for example, I could create the mailboxes thompson@ttgnet.com and barbara@ttgnet.com, and then configure my own mail package to POP directly from thompson@ttgnet.com and my wife's mail package to POP from barbara@ttgnet.com. The only limitation is that all virtual ttgnet.com mailboxes have the same password. Is this also the case with pair.com, or would I have to create and pay for individual mailboxes for each virtual ttgnet.com account name I wanted to POP from?

We do support extra POP boxes, but there is a charge attached. Each extra mailbox costs $1 per month per box. These boxes all have thier own POP login and password. You can request these extra boxes through our upgrade system.

So it appears that I won't have to buy a whole bunch of extra mailboxes if I move to pair.com. If I do that, I may just leave things as they are now, with Barbara's mail forwarding to our ISP account, or I may pay a buck a month for a real POP mailbox for her at pair.com. At least it's not going to cost me $20 or $25 a month more to have all the autoforwarded mailboxes I need.

* * * * *

Okay, it's about 17:05 EST, and I'm going to try publishing this again. If the past is any indication, at best it'll take ten or twenty minutes to publish and at worse I'll get a server tiime out.

Wednesday

Wednesday, February 3, 1999

And the publishing process completed normally. I started at 17:05, and it finished up a few minutes later, in about the usual time it takes me to publish in the morning.

With everything else that happened yesterday, I forgot to mention that the one of the fluorescents in the downstairs kitchen died, this the same day that Jim Wilson replaced the ballasts. I mailed him yesterday, and he mentioned that one of the tubes had a ring around it. We'll try replacing that tube today and see if that helps.

* * * * *

And I think I'll go back to publishing in the morning. It seems dumb to just leave stuff that's ready to go sitting there unpublished all day long. The ATX test-bed system is up and running. More details on the special reports page for it. That page is very much a work-in-process page, though. Once it's finished and cleaned up, I'll send Pournelle a copy to post on his web site. That's what the stubbed out notices at the top and bottom are all about, in case you were wondering.

* * * * *

Mail from last night:

This from Bolo [rabes04@ibm.net], regarding batteries for the Pentax Spotmatic:

Hello, I loved reading about your old Pentax Spotmatic on your web page. I have a Spotmatic that I got when my father passed away. Our 17 yr old son wants to use it for a photography class at school, so we dug it out. Still looks brand new, kinda proud of that. However, the only battery (for the light meter) that was in the case has absolutely NO markings on it. Would you have any idea what kind of battery this takes? Thanks in advance for any information you can provide!

Thanks for the kind words. As a matter of fact, you're the third person in the last couple months to ask exactly the same question (see my Daynotes for the weeks of 11/16 and 12/28). I took the battery out of my Spotmatic, and it's a standard Energizer EPX76 silver oxide hearing aid battery that you can buy in most drugstores. I also dug out my 1970-vintage copy of Herbert Keppler's The Honeywell Pentax Way, which specifies a Mallory RM400R or equivalent.

* * * * *

This from Gary M. Berg [Gary_Berg@ibm.net]:

In years past, it's seems like the "net" has often slowed down in January as all those people who bought PCs for the family for Christmas sign up with AOL, CompuServe, Prodigy, etc. Those one month free trials are irresistible, and I think these users contribute to an increase in net congestion. So it might be until the end of the month when the net starts to respond better. And, with the way people keep adding bandwidth to the net, they'll catch up with the added users soon too.

Good point. I hadn't thought about all those Christmas PCs, but that makes sense. And you're right about bandwidth. When I got started on the Internet back in 1987, the backbone itself ran at 56 KB/s, which was considered a "high speed" connection back then. Nowadays, they're deploying miles of fibre optic cable every day. I suspect it won't be long until little mom and pop ISPs have OC48 links.

* * * * *

And here's something from Chuck Waggoner [waggoner@gis.net] that might explain some of the slowdowns people have been experiencing:

I can report that, for about 2 weeks, up until last Thursday (28 Jan), I was experiencing incredible slowness with everything from logging on, to getting pages to load and email checked without timing out. This has been an occasional problem--after a few days of such behavior, it usually goes away. But after not even being able to log on after 5:00pm for almost a week, I was sure it was my ISP's fault.

During all these periods of problems, I had noticed, that occasionally the modem handshake would fall back to a 31.2k connection, and everything would work perfectly during that connection. I had wanted to try forcing a slower connection, but I couldn't find any of my old modem manuals with the AT commands--nor were they on USR's site, whose modem I have (forget what they now uselessly call a manual--it's really an advertisement for Web access). But aggravated beyond belief last Thursday, I contacted a buddy who knew the right info--and boy am I glad.

Since then, I've been forcing a 32k connection (anything slower does not enable error correction in my modem), and things have never worked so well. I haven't had a single failure of any subscription to download, or email to be checked, and those were formerly frequent--even when I thought times were good. In checking over the modem logs, both of those activities are happening in less than half the time they were when the modem was left to connect at its own discretion, which was usually between 44 and 52k. I can't explain this, and it seems counter-intuitive, but for others who are experiencing slowness, try this first. On my modem it's AT&N20 for a 32k forced connection. It's sure working faster for me when it's set slower!

That's a new one on me. I still use a USR v.Everything 33.6 modem, so 56 K speeds are outside my experience. Off the top of my head, the only thing I can think of that might explain what you're experiencing is a TCP/IP retry situation. That is, if your ISP has a whole bunch of high-speed modems in a large bank, it's possible that the cumulative throughput of all of them is swamping the ability of a router or other TCP/IP network component to keep up. Simplying grossly, your DCE-to-DCE rate may be, say, 50 K. If you're using that 50 K to deliver three packets in the time that the router can only process two, you're getting a lot of dropped packets and retries, which can easily translate into timeouts. However, I'd expect dropping the DCE speed to improve that situation only if you were connecting to a different physical network, which usually implies dialing a different phone number. So, I don't really know. I'll post this message. If anyone else has any ideas, I'd love to hear them.

Thursday

Thursday, February 4, 1999

Barbara had another bad experience while walking Duncan last night. This time, she twisted her ankle and fell. She had some minor scrapes on hands and knees, and a slightly sprained ankle. We applied ice to the sprain and Neosporin to the scrapes. But what worries me is that she hurt herself under the bottom rib. I think it's probably just a muscle pull, but she doesn't want to go to the hospital.

Our health insurance expired as of 1/31/99, and we're still waiting on Blue Cross to process our application. Tomorrow, I'm going to call the guy and tell him he has through the end of this week or he can forget it. We applied in mid-December, at which time they told us that we could expect an answer in about 30 days. Every time I call, they put it off longer. I'm tired of being jerked around. We're not really without health insurance, because we have 60 days to retroactively re-activate our old coverage under COBRA by paying the back premiums. I just hope that Barbara didn't avoid going to the hospital because of our health insurance situation.

Barbara is still hurting this morning. I asked her if she'd at least go to the doctor to get looked at, and she's thinking about it. She's a lot like me in that respect. Even before, she refused to go to the doctor unless there was a real problem.

* * * * *

Microsoft has apparently changed their position somewhat on Windows NT 4 and Y2K compliance. In the past, they've said that installing SP3 brings NT4 into Y2K compliance with minor exceptions. Now they're recommending that users install SP4 to ensure Y2K compliance. The complete story in the InfoWorld article Y2K nips NT users.

I've been running SP4 on one of my servers since shortly after it was released, and have not had any stability or performance problems or incompatibilities. I've heard few reports of problems with SP4, and those that I have heard about have been relatively minor. Conversely, SP4 fixes scores of bugs, some of them pretty major. You can view a complete list of the bugs fixed by SP4 (and earlier SPs) at http://www.nthelp.com/nt4bugs.htm.

I'd decided to run SP4 on one server for three months before committing to upgrading my entire network. Although the press of other business prevents me from doing so immediately, I plan to upgrade all my servers as soon as I can make the time to do so. I recommend that those of you running NT4 do the same. As always, it's worthwhile to do a complete backup before you apply the service pack, and then keep that backup available until you have satisfied yourself that SP4 works properly on your own system.

* * * * *

The following message from Jeff Erickson [erickson@innocent.com]. If anyone can help him any more than I was able to, please email him directly. My guess is that he has an Intel Seattle-1 SE440BX or a Seattle-2 SE440BX2, but I can't say that for sure based on the information he provided.

Hello Robert!

First off, let me congratulate you on an excellent web site. I've found tons of good reading and lots of useful information.

My reason for writing you is that I am in desperate need of a little technical help. You might not be able to help.. or even really care.. but I thought I'd ask anyway. I have a Gateway 450XL (Pentium II) with 128MB of SDRAM and a crap-load of good stuff. My problem arose as I tried to move my motherboard from my Gateway "mid-tower" case to a slightly larger tower case made by Enlight. The whole operation went fine up to the point that I went to connect the various wires leading to the front switches. It seems that Gateway uses a "Front Control Panel" set of pins to connect to their case. Enlight, and the rest of the modern world, uses individual wires leading to things such as the power switch, hdd light, speaker, etc. Gateway's Front Control Panel is basically a cluster of pins with no markings. I checked Gateway's web site for information on the pin-out for this Front Control Panel.. no luck. And, taking my existing case apart didn't help much as the wires are poorly marked and their connections to the front of the case are obscured by the front plastic grille. A phone call to Gateway Tech Support resulted in lots of "uhhhh.. I don't know" type responses.

Hillary Clinton and I are just sure that this is all part of a vast right-wing conspiracy to stop Gateway customers from switching cases. But, I'm still determined to figure out this pesky Front Control Panel pin-out. My question to you is this: In your experience with your Gateway system or in the contacts you have with others, have you ever come across a pin-out diagram for a Gateway Front Control Panel? For more background on the issue, you can download any of the Pentium II BX motherboard manuals off of Gateway's site. They all have a very pretty diagram of the Front Control Panel... but not a single pin-by-pin listing of what it does.

Hmm. Off the top of my head, I don't know what to tell you. As you've noted, there are two possibilities:

1. Figure out what the unlabeled motherboard pins are for by examining the motherboard itself. Although Gateway uses third-party motherboards, they may use different motherboards in the same nominal computer model, so the real problem is finding out exactly which motherboard you have, by manufacturer and model. If you can do that, you can visit the original manufacturer's web site and probably find the information you need. The last time I looked, Gateway was using a lot of Intel motherboards, so that may well be what you have. Gateway assigns their own internal ID numbers to motherboards. You can locate that number on your invoice or on the motherboard itself. Once you've done that, you should be able to find someone in Gateway tech support who can translate their internal ID to the real manufacturer and model. With that information, finding the connector pinouts shouldn't take long.

2. Figure out where the wires to the front panel connectors lead. I haven't looked at a Gateway case lately, so I'll assume that you've already done your best with this. But if you haven't popped the front bezel yet, doing that should expose the wires and connectors enough at least to tell which color wires lead where. That's probably the best bet.

There is, of course, a third possibility. Take your best guess. I've never damaged a motherboard by misconnecting front panel connectors, and I think it's theoretically impossible to do so. If you want to take a chance (obviously, you do so at your own risk), find out what the Gateway motherboard ID is, call up the picture of it, and compare it to some likely candidates like the Intel Seattle. If the component layouts look identical and the front panel connector block looks identical, chances are they are identical. My Intel Seattle board, for example, has the following connectors from left to right, where "X" is position with a pin and "B" is a position with no pin installed::

1. SPEAKER, 4 positions, XXBX

2. RESET, 2 positions, XX

3. PWR LED, 4 positions, BXBX

4. <one vacant postion>

5. HD LED, 4 positions, XXBX

6. <one vacant position>

7. INFRARED, 6 positions, XXXXBX

8. <one vacant position>

9. SLEEP, two positions, XX

10. PWR ON, two positions, XX

And, of course, the other problem you may run into is that the cable jumper blocks on the front panel wires on your new case may not match up with the pins on the motherboard. Sometimes, for example, you'll have a 4-position connector with position 2 blocked, and find that it needs to connect to a block of header pins that has all four pins present. That's easy enough to deal with by bending the surplus pin backward slightly so that it's out of the way (or by cutting it off, if you're really brave). What's worse (and it happens frequently) is that you have, for example, one four-position header-pin connector on a cable, with all four wires connected. You find that two of those wires need to connect to header pins on one side of the motherboard header pin array, and the other two wires need to connect to two pins on the other side.

* * * * *

And Scott Kitterman [Kitterma@erols.com] sent me, without comment, an interesting web page about a new method of predicting the stock market. I sent him following:

Hmm. I don't think I'll paste this one into my web page. I take it that you sent it to me as another example of the predictive value of collective opinion.

To which he replied:

Yes, that's correct. It struck we coming so close on the heels of your experiment.

Here's the URL we were talking about:

http://www.wired.com/news/news/business/story/17695.html

Friday

Friday, February 5, 1999

Important

Last week, we discussed whether problems existed with running Diskeeper 4 on a Windows NT computer with SP4 installed. We concluded that a serious problem that may cause data loss or corruption does exist, but the problem is not really with Diskeeper. The problem is a bug in the "repair" or "/r" option of the SP4 chkdsk program, and it can manifest even on an SP4 system that does not have Diskeeper installed. Also note that a similar problem may occur even under SP3 with volumes larger than 8.4 GB.

Unfortunately, running Diskeeper Directory Consolidation with the CHKDSK option enabled may cause this problem to rear its ugly head. Executive Software has posted updates for Diskeeper 3 and Diskeeper 4 that address the problem.

Didn't get a whole lot done yesterday, with my mother recovering from oral surgery and Barbara "whores de combat" (as a student once wrote in a paper I was checking). My mom seems to be doing better, and Barbara seems to be recovering. She's still sore, but it appears that it was a muscle pull after all. She never did go to the doctor.

I emailed the Blue Cross/Blue Shield guy, and he tells me that our application will be processed today. We'll see.

* * * * *

I've been having an exchange of mail with Tom Syroid about expanding system memory that I thought might be of general interest, so I'll include it here. It started with the following statement that Tom made on his web site:

Continuous "file thrashing" has become a rigor-du-jour for my system these days

Yep. My rule of thumb for NT is that if it ever hits the page file, that's too often. I don't know about there, but down here PC100 memory is about $1.50 US per MB. I standardized on 64 MB for NT client systems a couple of years ago, 128 MB a year ago, and now I think I'll start putting 256 MB in any new ones I build. More for serious servers, of course.

If you don't have at least 64 MB, you should run out and get some memory ASAP. If you have 64 MB and are doing much multitasking, you'll probably be surprised just how much difference going to even 96 MB will make.

Mmmm. High agreement on all your points. I have 64 MB, but since I've started "Web Designing", I've found myself habitually having 4, 5, 6... windows open at any given time and switching between as tasks dictate. I've held off buying any more ram for this system to date because (a) I plan to build a new system this year and it will definately have 256MB P-00 installed, (b) when I do, this system will be relegated to a games/Win95 system for my daughter, and (c) the max RAM capacity is only 128MB, and with what I have and what I want to get, I'll lose what I have currently installed in upgrading. Fiddle-faddle. Oh well. Looks like I'm going to go ahead and bite the bullet anyway. At least I'll be passing on a 'maxed' out machine to Danielle.

Umm. Before you do, I forgot one thing. You're running on a Pentium, right?

A lot of Pentium chipsets support more system memory than they can cache. For example, the current 430TX supports 256MB of system RAM, but can only cache 64 MB. If you have a chipset with this restriction, adding memory above 64 MB can actually slow the system down. On balance, I think that your system would probably run faster with more than 64 MB even if it had such a chipset, because even uncached RAM is much faster than the disk accesses that are occuring when you hit your page file.

Umm Indeed. Never heard of that one before...

I agree with your analysis of the pros and cons of this 64 limit as regards to pagefile access. But is there a way I can find out whether or not my chipset is indeed limited in cache size or not? My MB is a Compaq proprietry with a Triton HX chipset.

The 430HX is the best Pentium-class chipset that Intel ever made. It supports up to 512 MB of system RAM, all of which can be cached. About the only thing it lacks relative to newer chipsets is support for SDRAM. You'd mentioned a 128 MB limitation, which must be something Compaq did. One other possibility is that Compaq specified the limitation of 128 MB because of the amount of cache RAM installed. If so, you can also upgrade the cache RAM, which will allow you to support additional system RAM, fully cached.

One other thing to look for: the 430HX was Intel's butt-kicker Pentium chipset. It supports dual CPUs, and an awful lot of HX motherboards have an unused CPU socket. If that's the case, since you're running NT, you could install a second CPU in that box. It should match your current CPU in all respects. With two Pentium CPUs, you may find that your existing box runs faster than a lot of uniprocessor Pentium II boxes.

Although this started as a private exchange, we're discussing some stuff that may well interest others. Unless you object, I'm going to post our exchange.

No problem, post away...

I *wish* I had dual processor sockets. If I did, I'd certainly have 2 processors driving NT by now.

Curious point you make, though about the 512 MB max. My board came without an external cache chip and I finally found one (by Kingston) which my wife bought me for Christmas -- very spendy, but any improvement is progress with a P-100. I'm wondering aloud if the RAM max is specified by Compaq due to the normally absent cache, or if it is indeed hard-coded into the MB. At any rate, my plans include an upgrade to 128MB ASAP. My next machine, though, will very definately have provision for a second CPU. I may not fill the spot at first, but it will *definately* get filled eventually.

Without knowing anything about the motherboard, I can't really say. But I'd guess that your motherboard probably came with a certain amount of soldered-on cache and an empty socket for adding more cache. If that's the case, it may well be that the amount of soldered cache was adequate to cache only 128 MB, and so Compaq specified that as the system maximum.

Another possibility is that the SIMMs of the required size weren't available when the motherboard shipped. For example, if your motherboard has four SIMM sockets, although the chipset itself may have support for 64 MB SIMMs and 8 SIMM sockets (for a total of 512 MB), it may be that only 32 MB SIMMs were available in quantity when the motherboard shipped. Accordingly, Compaq may have listed maximum memory as 32 MB X 4, or 128 MB.

As far as your next machine, I think getting a dual-CPU capable system is a good idea. Running NT, you're likely to find that having two slower processors yields a faster machine than having one faster processor, and the total price may well be lower. Because of the way Intel prices the different CPU speeds, two mid-range processors (e.g. Pentium II/333) are also often cheaper than one of their current top-of-the-line CPUs (e.g. the Pentium II/450). Although using two Pentium II/333 CPUs doesn't literally give you the equivalent of a Pentium II/666, it does give you more power than the single 450.

As far as starting with one CPU, just be careful that the CPU you get will still be available when you want to add the second one. For example, if you start with one Pentium II/333, you need to keep a careful watch on the market to make sure that Pentium II/333's don't disappear before you get your second one. Of course, if your first CPU is a 100 MHz FSB one, you could always later add a second faster CPU and run both of them at the slower speed. Alternatively, you could also add a second faster one and run both it and the original at the faster speed. Pentium IIs are still overclockable, although that capability is going away soon, because Intel is putting PLL circuits on their CPUs to restrict them to running at their nominal speed.

A lot of people don't understand the dynamics of overclocking and CPU speeds. When Intel first ships a given model of CPU, their yields are relatively low, and they produce many units that won't run at the highest speed. So, they test each CPU as it comes off the line. When they first started producing Pentium IIs, for example, the few good ones that'd run at 333 got put in one bin. Those that wouldn't run at 333 but would run at 300 got put in that bin, etc. all the way down to 233.

But as their lines ramp up, their yields improve greatly, and the percentage of CPUs that will run at the fastest rated speed becomes much higher. Once they get all the kinks worked out, essentially all of the CPUs coming off the line will run at the fastest speed. But their pricing structure demands a premium for the fastest rated CPUs, and many people want the slower versions. So Intel arbitrarily pulls CPUs and marks them for the slower speeds, even though they would in fact run at the faster speeds.

What that means right now is that if you buy, say, a Pentium II/350, chances are excellent that it will run successfully at 450. At least until Intel starts putting the PLLs in to restrict each CPU to running no faster than its rated speed. And that fact in itself is telling. Because they have to set the PLL-restricted speed *before* they test the CPU. That means they're not concerned about wastage, because they're arbitrarily marking Pentium IIs at 350 that might have passed the test for 450.

* * * * *

This from Timothy Werth [timothy.werth@eds.com]:

Someone has probably told you this already but the link at the bottom of the page for the atx-testbed report points to the same page. The link: An extended version of this report, including benchmark tests, is available at http://www.ttgnet.com/Reports/atx-testbed.html

Is on page:

http://www.ttgnet.com/Reports/atx-testbed.html

Hope you have fun playing w/all of the new m/boards & hard drives. I look forward to reading your adventures.

Thanks, Tim. I was actually aware of it, and put a notice to that effect somewhere (I can't remember where). The reason that link points to its own page is that I am going to write the document in its entirety and post it at that URL. I'll then edit the document down to a much shorter length, and send it to Jerry Pournelle to post on his site. The link at the bottom is a way for people who want to read the additional information to jump over to my site. Same thing with the stuff at the top which welcomes visitors from other sites (except that part will stay for my site and go away for the version I send to Jerry).

Saturday

Saturday, February 6, 1999

This week has been a bad one from start to finish. But it's ending on a great note. My agent, David Rogelberg, mailed me yesterday to say that we finally have a signed contract from O'Reilly for the book that Jerry Pournelle and I are doing. There were a lot of things to get worked out. More than once, we thought we had everything worked out, only to find that one or another thing had popped up and we had to go back to the drawing board. But now, it's a go. I'm really looking forward to working with Jerry. I'm going to learn a lot.

* * * * *

This from Dave Farquhar [farquhar@lcms.org], responding to my comment about Windows NT 4 Setup restricting the size of volumes created during installation to 8 GB or smaller:

I've run into the same problem in the past with NT only using 8.4 GB of a large drive. The only solution to this I've ever found is to extract ATAPI.SYS from NT SP4, throw it on a disk, and when NT asks to set up mass storage devices, tell it to skip detection and feed it that disk. As long as your BIOS supports the big drives, NT will happily support it with this fix.

Going into Disk Administrator after installing SP4 also works, but if you like to partition your drives at installation time, this is the only way you're going to be able to do it. Here's a link that goes into more detail:

http://www.ntfaq.com/ntfaq/install47.html#install47

I've never tried that method, but it makes sense that it would work. I suspect that limitation won't affect most people, though, even those with large hard disks, because few people install NT in a huge partition. I generally create a 1 GB C: system partition for NT (or a 1 GB C: primary partition for Win9x and a 1 GB D: volume for NT, if I'm going to dual-boot) and then use Disk Administrator to partition the remainder of the drive. I suspect most people do the same. I'm just not comfortable having NT floating around in a huge partition with a bunch of other stuff.

The BIOS support Dave referred to is Extended INT13 Support. The major BIOS manufacturers--AMI, Award, and Phoenix--patched their core BIOS code in very late 1997 and very early 1998. If your BIOS predates that, you'll need to update your flash BIOS before you install a hard drive larger than 8.4 GB. Most drive manufacturers provide software workarounds for this problem, but I've never thought that using a device driver for something as fundamental as drive support was a good idea. Patch your BIOS instead.

And Dave follows up by saying:

I normally do the same thing. Monster NT boot partitions scare me, and they require some serious Partition Magic trickery anyway, since FAT partitions are limited to 2 GB and NT installation creates NTFS partitions by creating FAT partitions and converting to NTFS.

But since it's a little disheartening to have NT erroneously identify my drives, I tend to do this trick anyway. There was one occasion where I put a 10 GB DiamondMax Plus drive in a system, and for some reason I never figured out, the system BIOS autodetect misidentified the drive size. I ran into problems during the partitioning stage (for obvious reasons), but since both NT and the system were misidentifying the drive, I had a devil of a time tracking down the problem. When PartitionMagic also misidentified the drive, I went into the system setup and set the drive type to AUTO (overriding the values autodetect put in), rebooted, and the problem went away.

This was a really odd episode. I'm guessing that over the past five years I've built about 70 systems and upgraded about 500. I've only seen this particular problem once. But ever since, I've liked to have the OS work properly from square one with monster hard drives, just in case.

Speaking of partitioning, I wouldn't mind hearing some good discussion on partitioning. I'm in the habit of creating a small partition, changing NT's TMP and TEMP variables (Control Panel --> System --> Environment in NT; SET TMP=E:\TEMP and SET TEMP=E:\TEMP in AUTOEXEC.BAT on older MS systems) and pointing them at a subdirectory residing on that partition; moving Web browser cache into another subdirectory on that partition; and then storing programs and data separately (data preferably on a network drive, but on standalones, on a separate partition), on the theory that programs, left to their own, aren't terribly likely to get fragmented. Browser cache and temp files, on the other hand, by their very nature are the most likely to get fragmented, but at least by quarantining them, they're not fragmenting my important stuff.

Some programs don't behave and use C:\WINDOWS\TEMP regardless of what I do, but the majority seem to follow whatever I set those environment variables to use. Ever since I started doing this, I've found I don't need to defragment nearly as often.

The other advantage I've seen to partitioning is that sometimes the OS (especially DOS-based OSs like Win95/98, but I've seen NT and OS/2 do this too on a couple of occasions) will corrupt data in its own partition after an especially nasty system crash, but usually the other partitions will survive. Isolating the OS to its own partition provides some damage control, and makes disaster recovery with Drive Image or Ghost really nice.

Hmm. I've dealt with a lot of disk drives over the years, and I don't recall ever having a system mis-identify a drive. As a matter of fact, I'm not even sure how that could happen. I think it was ATA-2 that implemented the IDENTIFY DRIVE command. When the ATA interface issues the IDENTIFY DRIVE command, the embedded drive controller returns the string stored on ROM that contains the drive model, CHS counts, etc. I don't doubt that it happened to you, but it would be interesting to understand the sequence of events that allowed it to happen.

I don't go as far in segregating stuff as you do. I usually create a 1 GB or 2 GB partition for the OS and applications and put the data on a network drive. Since the drives in my systems run from 3.1 GB up to 10+ GB, that leaves me with a lot of unused space on workstation drives. I take advantage of that space by creating dedicated volumes for various stuff. For example, I use one of my workstations as a distribution server. It has a "left over" 6 GB partition filled with copies of distribution CDs. It's shared with the network, so I can install anything any time without tracking down the original CD. I put other leftover partitions to similar tasks. I have a partition dedicated on each of two machines to storing the backups that I do via batch file several times a day. I have plans for one huge leftover partition to store MP3 audio files. That way, I'll be able to play my music from a network server instead of loading the original CDs. I also find the leftover space convenient for doing snapshot backups with DriveImage.

As far as pointing standard Windows directories to other locations, I made an executive decision a long time ago not to do that. As you point out, everything usually works fine, but you'll inevitably run into applications written by stupid programmers who hard-coded directory names (and even volume ID letters) into their programs. So, I just let Windows point where it wants to. I don't worry much about fragmentation. I have Diskeeper 4 set to auto-defrag at 3:00 a.m. each morning, and my fragmentation level seldom rises above 1% on any volume.

And Dave follows up again:

The only cause that I can think of for the problem is a flub-up at the CMOS level. I don't remember if there were any CMOS checksum errors around that time or not. I know that system's original motherboard was troublesome, but I don't remember now if I put the drive in before the motherboard swap or after.

It was certainly odd -- I've only seen it once time out of hundreds of systems and I'll be very surprised if I ever see anything like it again. But I tend to be over-cautious, if anything. Computers are just too valuable, from a practical if not from a monetary standpoint, to be reckless with them.

I do know that once I got the Maxtor drive working properly, it kept right on working properly, not to mention quickly. I can't say enough about the DiamondMax Plus drives. A Seagate Cheetah is a little bit faster, but the Cheetah plus a decent host adapter will cost three times as much and give you (at best) 20% better performance.

I agree that a problem with CMOS seems the only explanation. And I share your high opinion of the Maxtor DiamondMax Plus drives. As you say, the Seagate Cheetah is marginally faster, but it's also a 10,000 RPM SCSI/Fibre Channel drive intended for departmental and enterprise servers. In my opinion, you can't beat the Maxtor DiamondMax Plus drives for individual PCs and small servers.

* * * * *

This from Shawn Wallbridge [shawnw@elections.mb.ca]:

You Said....

Of course, if your first CPU is a 100 MHz FSB one, you could always later add a second faster CPU and run both of them at the slower speed. Alternatively, you could also add a second faster one and run both it and the original at the faster speed.

Actually this is not possible unless you buy a 300 and a 450. Intel locks the multiplier on their chips. So a 4.5x 300 will run at 450 but you couldn't take a 350 and run it at 450 (without overclocking the bus). I have a PII 300 SL2W8 that has the 4.4ns cache chips on it, so I run it at 450 and it is perfectly stable. Intel has not officially said that they are planning to clock-lock any CPU's yet but I have heard rumors, that they are planning to. This would be the end of overclocking without physically modifying the chips.

I just purchased a Celeron 300A and I plan to run it at 450. This has to be the biggest bang for your buck ever. I paid $140CN and I have a machine that is faster than a PII450. This will give me two PII450 class machines to play with. If you are looking at overclocking a PII the only motherboard to buy is the Abit BH6. It allows you to adjust the core CPU voltage in .5v increments.

Arggghhh. You're right, of course. One of the hazards of using modern tools like word processors is that they've made it *too* easy to cut, paste, and edit. Then there's the fact that I'm so busy that sometimes I don't pay close enough attention to what I'm doing. Thanks for the catch.

As far as Intel locking CPUs, I believe that it's now reality. It may count as rumor, because I haven't verified it myself, but I've been told (unofficially) that Celerons now shipping into the channel are locked. I have a Celeron/333 sitting here that I haven't done anything with. It runs a 66 MHz FSB with a 5.0 multiplier. I may plop it into this EPoX EP-BXT motherboard and set the FSB to 100 MHz to see what happens when you run a Celeron at 500 MHz.

The Abit BH6 is indeed a good board, but I suspect many readers won't understand your reference to voltage. For their benefit, I'll explain that CPUs (indeed any chip) "like" higher voltages better than lower ones (within reason), and lower temperatures better than higher ones. For example, years ago we used to test DIP RAM chips. Our tester allowed us to vary the voltage from the nominal 5.0V and test how reliable the chips were at various access speeds. If we found a chip that was reliable for 60 ns applications when running at 5.0V and normal room temperature and then re-tested that chip at 4.75V and a higher temperature, we'd typically find that it's reliable access speed became something like 125 ns. Same deal in reverse. A 120 ns chip running at 5.25V with an ice-cube sitting on top of it (literally) would test reliably at much faster speeds.

The same things hold true for processors. When manufacturers rate speed on CPUs, what they're really rating is how fast that chip can reliably run at nominal voltage and temperature. If you put a cryogenic cooler on a CPU and boost the voltage slightly, you can run it much faster than its rating. Then, of course, there's the small matter that the actual reliable speed of a chip may be much higher than its rated speed, simply because the manufacturer needed some "slower" chips to meet market demand. That's the case with the Celeron you have. It never really was a 300 MHz CPU. It was a 450 MHz CPU with a 300 MHz label on it.

And Shawn adds:

I have been keeping up with the clock-locking. I have read reviews of the Celeron 400 and the PIII500. The Celeron 400 would boot at 600 but would crash almost immediately. The PIII would also boot at 620 (5x124) but wouldn't run stable. The PIII was a engineering sample, but the Celeron was production. I read at www.tomshardware.com that his sources at Intel denied the clock-locking. I would think that the engineering sample would contain the clock-locking. Either way I have two PII 450 class processors for about $500CN. I am happy for now.

Dunno. But I should have some current production samples of the Pentium II/450, Celeron PPGA, and Pentium III coming in the next month or so, so I'll check them out.

And I agree that the Celeron offers by far the most bang for the buck for mainstream use. Relative to the Pentium II, about its only real drawback is its lack of support for SMP. But if you're building single processor boxes, the Celeron's 128 KB of full CPU speed cache doesn't give up much, if anything, to the Pentium II's 512 KB of cache running at half CPU speed.

My guess is that the Pentium II will continue to sell for a while, at least while the Pentium III is priced high. Once the Pentium III prices start to drop, though, I'd bet the Pentium II will disappear, leaving the Celeron as their mainstream CPU and the Pentium III for servers and SMP clients.

* * * * *

This from Dan Bowman [DanBowman@worldnet.att.net]:

Thank you for posting the story about memory links in men and women. The staff here absolutely loved it. I think that everyone (regardless of gender) could relate some aspect of the tale.

Thanks.

* * * * *

This from Tim Werth [twerth@kcnet.com]:

http://www.tomshardware.com/news/index.html

Since you have been discussing changing your web hosting service I thought you would find it interesting that pair networks does the webhosting for Tom's Hardware Guide. I believe Tom's gets A LOT of hits so a word from Fredi recommending pair networks says something. Talk to you later.

Yep, there's no doubt that pair Networks has superb connectivity. My problems with BigBiz have pretty much gone away, though, and inertia says it's easier to stick where I am for now. If I move, I'll almost certainly move to pair Networks, but I've got so many other balls in the air right now that moving is lower on the priority list than a lot of other stuff.

* * * * *

This from Ken Scott [kscott@pcisys.net] regarding main memory cacheability with Intel chipsets:

Regarding your discussion with Tom Syroid about motherboards and the amount of RAM that can be cached --

From everything that I have seen and read, a TX chipset can cache as much RAM as is installed on your system. The chipsets that couldn't were HX, FX and earlier sets. HX boards could cache more than 64MB of RAM if you installed a Tag RAM chip on the motherboard. These typically need to come from the motherboard manufacturer, and not all motherboards supported it.

Just wanted to clear that up a little

I think you've gotten some wires crossed somewhere. The original Triton chipset (82430FX) supported a maximum of 128 MB system RAM, of which it could cache 64 MB. Intel introduced the 430HX and 430VX as a "family", with the HX being their "performance" chipset and the VX their "economy" chipset. The HX supports a maximum of 512 MB system RAM, all of which can be cached. As you say, not all motherboards have the necessary amount of SRAM on board to cache the full amount, but most allow it to be added. The 430VX supports a maximum of 128 MB system RAM, of which only 64 MB can be cached. The 430TX (Intel's current Pentium-class chipset) supports a maximum of 256 MB system RAM, of which only 64 MB can be cached.

Ken's message got me to questioning what I thought I knew, so I bopped on over to:

http://developer.intel.com/design/chipsets/mature/index.htm

which confirms what I thought.

Sunday

Sunday, February 7, 1999

Barbara is mostly recovered, but I suggested that we not bother doing the weekly cleaning today. She insists on doing it, although we'll do a lighter job today than usual.

* * * * *

I've gotten several messages from people who are angry to discover that their chipsets won't cache the full amount of system memory. Some of them seem to think that Intel somehow misled them or produced an inferior product. That's not the case, however. In the first place, it's not just Intel that produced chipsets that won't cache all of the installed memory. Many chipset manufacturers did the same, and a 64 MB limit on cacheable RAM was pretty standard.

To understand why, think back to what you were paying for memory a few years ago. Fifty bucks a megabyte, more or less, right? Well, at $50/MB, 64 MB of RAM comes to something over $3,000. Chipset manufacturers assumed, reasonably enough at the time, that no one was likely to install more than $3,000 worth of memory in a low-end PC. And they did make provision for people who did need to install a lot of memory. Intel's high-end chipset of the time, the 430HX, supported 512 MB of system memory, all of which could be cached. Then, too, the chipset manufacturers never made a secret of those cacheability limits. Their spec sheets are right out there for anyone to look at. So being angry with them for products they produced several years ago is like being upset that Office 97 won't fit on that 10 MB hard disk you pulled from your old XT.

* * * * *

More on the Diskeeper issue. This message from Diskeeper tech support showed up in my in-box yesterday afternoon. I've removed the salutation and material at the beginning and end of the message that talks about being removed from their mailing list. The key section follows:

Free Update improvements have been made in Diskeeper 3.0 and Diskeeper 4.0. This email is being sent to notify you of these changes and to let you know you can download these updates for free at our web site. To download go to: www.diskeeper.com/download/

Improvements in these two versions: When the CHKDSK option is selected in the boot-time defragmentation dialog box, it is now only run once before the boot-time defragmentation process, greatly speeding up the entire operation.

Diskeeper 4 Build 202 (Intel Only) is for Diskeeper 4.0 workstation and/or server owners.

Diskeeper 3.0 Build 178 (Intel Only) is for Diskeeper 3.0 workstation and/or server owners.

If you have not upgraded from 3.0 to Diskeeper 4.0, try it now to check out the latest in increased speed and performance. If you want to try Diskeeper 4 (build 202) go to www.diskeeper.com/trialware/dkguest.htm

To which I replied:

I notice that your message makes no mention of data corruption that can occur when running Diskeeper with Windows NT 4.0 with Service Pack 4 applied, or when running Windows NT 4.0 with Service Pack 3 applied on hard disks larger than 8.4 GB.

I assume that the primary purpose of these updates is to solve the data corruption problems (which I understand are caused by bugs in Windows NT rather than in Diskeeper). If this is the case, it seems to me that you owe it to your users to mention the existence of that problem and to encourage them to apply the patch before using Diskeeper again.

Can you comment, please?

* * * * *

This followup from Ken Scott [kscott@pcisys.net] :

I also found the original sources that I had seen, and lo and behold, I didn't remember what I thought I did. That's what I get for writing from memory! It also reminds me that my main machine has a TX chipset and 96MB of RAM installed. Maybe I'll get to swapping some of that RAM to a deserving machine.

Thanks for listening,

Yep, we all get burned on occasion when working from memory, but what's the alternative? If we took the time to check everything we were pretty sure we knew was true, we'd never get anything done. As far as downgrading your memory, there's a good chance that doing so will improve system performance, but it'd probably be worthwhile running some benchmarks before and after.

* * * * *

This from Jurgen Beck [jbeck@sekd.com] regarding the new Travan TR-5 NS20 tape drives:

I read with interest your article about the latest Travan tape drives. Great article and suggestions!

You mentioned that the NS20 tape drives can be purchased for around $350.00. Would you happen to have a good source for them. The cheapest I have been able to find the Tecmar NS20 drive is $470.00 for an internal with the software. Is that the lowest? Thanks for your help!

I wrote that article well in advance of its expected publication date, so I had to guess at prices. As it turns out, though, I guessed pretty well. I just jumped over to http://www.shopper.com/cgi-bin/nph-find?tag=st.sh.fd.se&a0=NS20 where I found that Tandberg is selling a bare SCSI NS20 drive for $324, and Aiwa has one for $275. If you're interested in an ATAPI version, I see that the Seagate TapeStor NS20 (http://www.shopper.com/cgi-bin/nph-sort2?a0=289448&a1=4) is showing up for about $350 from the least expensive places.