Building the standard workstation

D. J. Bernstein's 2003.06.27 standard workstation was built.

Although the purchasing of parts began 7/21/03, because the author is middle aged with a job, construction of the workstation was not completed until 9/10/03, after the spec for the 2003.08.04 standard workstation [page moved] came out. Below, some pitfalls and reflections are cataloged.

Why Build a standard workstation?

The author has previously built four PCs. The first PC, a white box AT mini-tower, was handed down to him in '98. The motherboard was upgraded right away and eventually most other components were changed. Then next three PC's were built starting from bare-bones kits.

The first, built for the author's daughter, used the drives from an old Compaq with a new case, PS, motherboard, CPU and memory. This machine worked OK for a while but the motherboard cracked when swapping the sound cable from the old CD to the one from a new CD-R and had to be replaced.

The second bare-bones was built to run Plan 9. When it arrived, it wouldn't boot, but seating and un-seating the power connectors to the motherboard and removing and reinstalling the CPU got it going. It has worked fine ever since although the author abandoned Plan 9 when he figured out that running it wasn't going to get him a job at Murray Hill. Bare-bones number two now runs Windows 98 for the author's wife.

Bare-bones number three was built as a voice-mail server for a telecom client. The server was very unstable. The client's money was refunded and the server, minus voice cards, ended up under the author's desk. As the author's main computer, the mini tower, aged, he turned it into a Samba server and set up the reject server as a Linux work station. All went well for about 10 months, but then bare-bones number three started having random crashes that continued after a re-installation of Slackware. This box was not worth messing with. Because ISA slots were needed for the voice cards, one of the last AT cases on the market had been purchased. The case requited four hands to close and was made of paper thin metal. The author was in the market for a new computer.

The component at a time upgraded mini AT has been the most stable of these machines. The author suspects that shipping motherboards with fan and heat sink installed causes undue strain.

When Bernstein's Advice for computer buyers was stumbled upon, the author already had a preference for AMD CPU's, Gigabyte motherboards, Antec cases, and was using that $11 D-Link NIC in the Samba server. Bernstein seemed to have a taste for components similar to the author's and a decision was made to build the standard workstation, following the specification as closely as possible.

Pitfalls

ECC Memory

Between the time the spec had come out and the beginning of the shopping process, only four weeks later, the specified motherboard and in fact any quality ECC single processor motherboard had become scarce. Bernstein's 2003.08.04 spec includes a dual processor ECC motherboard. The author found a stale, overpriced GA-7DXE, v. 2.0 at mbx.com for $73.79, including shipping and ordered an AMD Athlon Thoroughbred 2000+ ($66) as he was concerned that the 2400+ would be too fast for this motherboard.

Moving Slackware

The author's successful plan was to use his old 20 Gig. hard-drive with an existing Slackware 9.0 install. He imagined that he would only have to change the modules for the graphics card, NIC and on-board sound. There were two unexpected problems.

Switching the mouse from the serial port on the old box to the PS/2 port on the standard workstation was troublesome, as the name of the PS/2 device, /dev/psaux was not known to the author. The PS/2 port would not set up properly unless the computer booted with the mouse plugged in and sleep was lost figuring this out.

Initially monitor power management didn't work under X. The new XF86Config, generated with xf86config, lacked an Option "dpms" line in the device section for the Sapphire 128 Pro. It took a while to find this. The XF86Config device section for the $25 Sapphire looks like this:

Section "Device"
    Identifier  "sapphire128pro"
    Driver      "r128"
    Option      "dpms"
EndSection

Other than this, there have been NO migration problems.

A line added to /etc/rc.d/rc.netdevice loads the module for the $11 D-Link DFE-530TX+ NIC:

/sbin/modprobe 8139too

This line was added to /etc/rc.d/rc.modules to load the module for the on-board sound:

/sbin/modprobe via82cxxx_audio

That Round IDE Cable

Bernstein specifies a single $15 plus shipping round hard drive cable. Although this is not the author's usual practice, a decision was made to set up the drives using only this single cable. The CD-R was installed in the top bay and the hard-drive in the "cage", not in a floppy bay. The cable wouldn't reach both drives, so the ribbon provided with the motherboard was used for the CD-R.

Actual Cost

The parts for the author's standard workstation cost $366.52, using the hard-drive and CD-R salvaged from bare-bones number three. Overruns were $15 on the motherboard and and $8 on the ECC memory. A savings of $19 was realized on the slower CPU.

The ECC DDR price given in the spec may be the price for non-ECC memory as there was an exact match.

Afterthoughts

The full tower case, a first for the author, both in quality and size, is very nice to work with. The case is easy to open; the side panel can be opened and closed without tools. Heavy gauge sheet metal is used in constructing the case and the corners are braced, resulting in a rigid case that will prevent flexing of the motherboard under normal conditions. The real estate this case takes up in the author's small office has caused some regret that a mid-tower case of similar quality was not purchased.

ECC memory may be a dead horse. The author would not consider going to 2 CPU's, as the current standard workstation does, just for this feature. He would instead forgo ECC and go for a modern motherboard with USB 2.0 and support for a faster CPU.

Shopping was by far the most time consuming part of building the standard workstation, taking about five evenings. Assembly took under an hour, and accommodating Slackware to the new box took about a night, but could be done again in an hour or so. The author suggests picking one or two suppliers with good service and parts fairly close to what is wanted. The savings in time would outweigh the few extra dollars spent.

Experiences with the suppliers Bernstein recommends, mostly new to the author, were positive. Please note that Googlegear has changed their name to ZipZoomfly.com abruptly in mid September, 2003. For a different spin than that Xtraplus Corp.'s press releases put on this, visit http://arbiter.wipo.int/domains/decisions/html/2001/d2001-0125.html.

Conclusion

The author has been using the workstation for a couple of weeks and so far found it stable and fast. He learned a lot and had fun building it. Thanks to Bernstein for designing the standard workstation.

Thu Oct 2 11:19:51 EDT 2003