WeekEnd Feature: The perils of upgrading.

by Ian Thompson, CCSP Staff Editor
March 6, 2004

We're constantly being told to keep our systems up-to-date. Up-to-date is good. Up-to-date means all those niggling 'features' are taken care of. Up-to-date gives us the confidence that we are running our PCs at their peak.

Well, welcome to my systems, where the normal rules seem to have sidestepped reality - just a touch.

So, up-to-date is bad?

Let me put it this way - there's leading edge, and there's bleeding edge. It is possible to be too far into the new stuff that it hurts. This was the point made to me as I considered the specification for last year's upgrade to the school's network. Let me just highlight what we did…

Any system slips from the lofty heights of 'current' to somewhere just south of bearable. Depending on circumstances, this can take anything from 6 months to many years - it's all dependant on the particular client needs, and more likely, the cash available at any one time. What one person considers fairly nippy, another will think is only slightly quicker than a legless dog chasing sticks. That 64MB P100 makes a fine word processor, and there's nothing wrong with a Psion PDA, but they're not exactly la mode plus élevée…

Our school used to run a Curriculum Network based on two PIII-550 servers, single CPUs with 128MB RAM and about 70GB RAID-5 storage. It had a hub-based unmanaged star network running over Cat5 copper - tops of about 100Mbit. Now, this was considered to be adequate at the time it was installed in 1998. Certainly, given that we had about 180 workstations (maybe a few more) that were either older freestanding ones or bought in new, this didn't really pose a problem. Compared to the smaller Administrative Network, running over a 10Mbit BNC'ed loop, it was a huge advance, especially given the ease at which connections could be modified to suit. We'd had about 400 sockets installed, and switching to patch about 2/3 of these through to the servers - it was ahead of the needs at the time.

By the time we were considering our application to become a Specialist Technology College, in 2002, this was beginning to look embarrassing. For example, our PDC server had completely failed early into the 2001 academic year (unfortunately just after I'd taken over the reins from the previous manager). I guess someone somewhere could work out the odds of having both of the redundant PSUs in it die at the same time, but I'm guessing that the odds, were they applied to a horserace, would dwell far into the 'outsider' category - the sort of things that punters put spare change on, just in case it romps home and they can retire on the proceeds. I'd done my best to eke out the performance, upping the RAM to 640MB in each, rearranging the cabinets and so on, but clearly only so much could be achieved. Despite this, no provision was made in the plans for our enhanced status to carry out a review and comprehensive upgrade of the infrastructure. It took some subtle manoeuvring and gamesmanship to have the situation examined, but eventually the situation rose to Okay, you can have £10K. Which would have been fine, except we were looking at a system approaching 330 PCs at the time, and more were on the cards. A bit more ducking and diving, and the final verdict was Okay, we can afford £150k - much more like it!


Sounds like a good case for upgrading, if you ask me…

Oh, I don't deny it. However, we went from an NT4 system with minimal centralised services and not much management to one based on multiple multi-processor servers running Windows Server 2003. Our 'bare minimum' has now sprouted two identical mirrored servers, a third that is scheduled to run the intranet services, the old pair reassigned as Print and Mail servers, and a dedicated hardware CD/DVD server by Avantis. The previously separated Admin system is going to join the fun shortly when it is linked to allow delivery of student registration information to staff, with one really creaky server, one fairly decent one and a third that will run Server 2003 Enterprise to handle the RADIUS stuff on our wireless system. There, those are the 'nine servers' mentioned in my biog.

Ah yes - the WiFi stuff. You know, there's a case where the constant upgrade cycle has not helped at all. Every time we hit a snag (just trying to do what the boxes claimed at the time), we managed one work-around or another, and have so far had four firmware upgrades (at least one of which was written for us as a direct result of our experience). I won't say who the manufacturer is because they've just agreed to replace the original access points with a newer, clearly more heavyweight version and I don't want to risk the chance they read this and pull out. Nothing wrong with the PC Cards they produced, or the POE (power-over-Ethernet) equipment at all - they give solid performances - but the AP's are, well, obstinate. That's the nicest way of putting it. It's not without some wry comments (did I say 'wry'? I meant 'blue') that we've noticed that the entire range has now been pulled from production, wiped from the company website, and all support ceased. Even before this, the declared specs were altered to remove the claim that it could do what we were trying. Still, that's another article in the making…

Our system is now tootling along at 1Gbit, largely over fibre, (one spur off is 100Mbit copper, but this will change by September), with a spanning tree, managed switches and a single manufacturer base. Read back in my archive and you'll see that the off-site link is about to get a serious boost as well. With the exception of the WiFi, it has been largely painless. This is especially remarkable because we were, to my knowledge, the earliest adopters of the newest server OS in all the schools in our city. Even some of the big companies were just considering it. I haven't heard Gartner's usual cry of Wait till SP1 before using it on any mission-critical blah-de-blah-de-blah, but I'm sure they've said it about Server 2003 just like they have about everything else.

So what's wrong?

It's this - effectively we're beta-testing stuff, and that's not on, considering we paid for a working system. No criticism of the installers at all, but especially the wireless side where the last-but-one solution was to try beta drivers for the PC Cards (not the AP's, where the problem clearly lies - AiroPeek never lies, but perhaps that's another article as well). We were also ahead of the city's ICT support systems, which have since caught up.

It's often said that systems need testing before being relied on heavily. Clearly this is followed rigidly; otherwise we'd have problems of critical banking systems being virus-ridden within weeks of being activated, or pin-and-chip payment cards that don't work in the new authorisation systems, and the old cards are no longer accepted. (What? Oh, I've already covered those points? And they didn't figure these things out before rollout? Okay…)

And yet, the first reaction by helpdesks and their ilk is often to enquire if the latest drivers are installed since these will solve precisely the problem were having, whatever that may be. Allegedly. Who knows? But give it a go any way because it gets the punter off the line and, well, the coffee's going cold.

Rant Alert!

Damn right. It's not just work that suffers from this issue. For example, I've just regained the use of my Briefcase archives. This is serious, because similar troubles befell other users of this speedy archiving tool. If it is the only method of backing up the changes of many peoples' work rather than the entire file store, then one or two upgrades have played havoc with it. In my experience, upgrading the JET database engine has been the culprit, as follows:

Version 6 seemed fine - maybe it's because I'd installed Office 2k at about the same time, then archived things onto Zip disk using Briefcase. Then, MS Windows Update posted a version upgrade for the JET engine and I lost the ability to add files to the Briefcase. This wasn't because Briefcase itself was altered, but because Access couldn't cope with some of the file structures correctly, giving a strangely uninformative message about using my original application to upgrade the archive. Recently, it went to JET v8, so I decided to swap the turnip for the thinking head and get to grips with things. The path from this upgraded database engine lead through Access, then through the Replication feature (that allows a master database to update numerous slave ones wherever they may be) and eventually to the enigmatic instruction that apparently implied that there was, somewhere, a tool designed to update Briefcase archives. This may have stumped many users - but only if they dared to use Access in the first place, so we're safe then since it's likely to be the least used part of Office in the SoHo user arena. However, to lose a backup system like this is worrying.


It can't all be bad....

Then there's the perennial favourite - graphics drivers. Take it from me - avoid every other nVidia release. I've normally tried to eke out every scrap of performance from my graphics cards (being tight-fisted), but it seemed that alternating driver releases either resulted in a black screen or generally bizarre performance in one area or another (usually when pushing the card hard). The last round was silly - I have just scooped an FX5900 card that recognised the drivers straight away from the previous incumbent (a GeForce 2MX) running on Detonator 53.04. I'd managed to get this up to 2827 3D-2001 marks - not a rocket ship, but a gradual improvement of about 40%. The FX unit clocked a relatively measly 7767, although on the newer DX9-compatible 3Dmark2003 tests it blitzed the old card by 4084 to 121 (really). I then popped version 56.55 on - the next release, despite the ditzy numbering - and got a black screen whenever things got a bit hairy on either 2001 or 2003 tests. Back on 53.04 and things run fine again. Then I was told about NVHardPage - no more regediting to get the CoolBits tweak! The tweaked card is running fine at 430MHz core/820Mhz RAM.

The same with my CPU - I run an apparently unworkable setup of an Abit KT7-RAID (KT133) mobo that declares an Athlon 1400 is the top speed it can understand, but the CPU is actually an Athlon XP2200+ B-core Thoroughbred chip, which should run at native 1.8GHz, but the mobo instantly overclocked to 2GHz, with a rating of pr2670. And from what I've read, I'm not the only one to have 'discovered' this cheap upgrade.

Maybe I'm too harsh - hardware upgrades seem to be fine; it's just the software upgrades that spoil things. One day, I may move from Win98SE on this PC, but given my record that will not be a smooth journey at all…




by Ian Thompson ComputerCops Staff Editor



Ian Thompson is a Network Manager of a 500-PC, 9-server, 1700-user school network and is an ICT teacher at a UK high school near the city of Leeds. He has written articles for the Hutchinson Encyclopedia, plus many resources in support of teaching ICT in the UK schools' National Curriculum.


Copyright © Ian Thompson All Rights Reserved 2004.