Peter's Computer News

April 11, 1997

Dual Pentium motherboard

Always looking for new things to try, I was fascinated with the idea of a dual-processor system. Since I was running NT, and NT supports multi-processing, it seemed the thing to do. The price of a dual-processor board had come down quite a bit, but the final motivation was a deal I could not refuse. A vendor, to whom I had referred a fair amount of business, wanted to thank me. He asked what I would like as a bonus. As it turned out, a motherboard with dual-Pentiums was within reason. They were only 75MHz Pentiums to stay "within reason", but the price was right.

I was eager to try the new board, but right off there was a problem. While the board included an EISA bus, which was not of use to me (EISA accessory boards are expensive), the motherboard did not have an EIDE, floppy, or I/O controller on it. Wanting good performance, I sought a PCI EIDE controller. Turns out they are hard to find. Most motherboards now have the controller built-in resulting in a very small market for add-on boards. Being stubborn, I was not going to settle for an ISA controller and finally found a Tyan controller through an Internet search.

With the arrival of the controller board, I was ready to assemble my prize system. It went together fairly well, but configuring the motherboard was a new experience. Did you know that the EISA bus requires configuration? I didn't. There is a floppy with configuration utilities that comes with the motherboard. The utility ran fine, but it asked me to set values that I had no idea what they should be. After poking around a bit, I got the feel for it and the computer booted. After some minor adjustments to the NT settings, I was running a dual-processor system. Just wait till I tell all my friends!

Now it was time for benchmark tests. I decided to concentrate on the software operations that ran the slowest on my old system. To run the tests, I measured time to load as well as conversion of graphic files for two image-cataloging programs. To get a reference point, I first tried the test runs with a single processor active and the computer configuration similar to what I had been running on my previous Pentium system. I also had access to a pair of 133 MHz Pentiums to try in the system. The best way to present the results seems to be in the form of a table. Here is the tabulated summary:

In reviewing the data, it appeared to me that there was no gain from having two processors for the programs that I was running. This was true even when two conversion processes were running at the same time. In some cases, having two processors actually made things slower (just a bit, but slower). The bigger gain was from having a faster CPU and adding more RAM. This being the case, I ended up trading the dual Pentium board in for a single Pentium 166 MHz motherboard with 512k cache and added another 32 MB of RAM. The result was that I had a better performing machine, although I did give up being able to brag about having a dual processor computer.

I ran some of the same tests on my new

Dual IDE drives in NT stripe set

One of the nice features of NT is that it provides a very flexible configuration of hard disks. You can assign any drive letter you want to a drive. It is also possible to join two separate had disks to appear as a single drive. The feature that caught my attention, though, was the ability to set up what NT calls a "stripe set." This is actually a software enabled RAID configuration. In theory, splitting data among two hard disks for read and write should double your throughput. Since NT uses virtual memory and does a lot of disk accesses, this seemed to be the thing to do.

With all the "bloatware" that I keep installing, my hard disk space keeps shrinking. In addition to the new 166 MHz Pentium board mentioned earlier, I purchased a second Seagate ST32140A IDE drive. Here was my plan: I had two matching 2 GB IDE drives, and there were two IDE controllers on the motherboard; a natural stripe set. There was a slight problem to be solved right away. NT cannot boot from a stripe set! I also had a 1 GB SCSI drive in the system. However, past experience showed that I could not boot from the SCSI if an IDE drive was present. This was the motivation to solve the SCSI boot problem. Of course I did solve it. See my March '97 newsletter for that article.

Having configured the drives and set up the stripe set, I was ready for the high-speed disk access. After starting the computer, I did not notice any obvious improvement in drive performance. Some tests were in order. I took some large directories (almost 100 MB) and copied them between drives. I copied Stripe to stripe, stripe to SCSI, SCSI to SCSI, SCSI to stripe. Interesting results. Although the system was running RAID on the IDE drives, and the Seagate IDE drives had a faster access speed than my older SCSI, the SCSI outperformed the stripe set! The difference was not much, but it was definitely faster. Later testing of a non-striped IDE showed that the stripe configuration did not result in any speed advantage. Perhaps with SCSI devices it is better, but with IDE drives, don't bother.

Since I do a lot of beta testing, it was not long before I managed to corrupt my NT system. The only solution was to re-install NT. Surprise, a new NT installation it will not recognize the stripe set! What about my data? Fortunately, I had a tape backup of my NT installation. By restoring the NT system from tape to the boot drive, I was once again able to access the stripe set drives. Lesson leaned. Since there was no speed advantage, and a whole lot of danger involved, I removed the stripe set and am now using the two drives as separate devices.

Home