I decided to stick with two tests: looking just at the actual disk performance under RAID 0, and what affect it would have on a benchmark that is intended to reflect current ‘Real World’ performance. For actual disk performance, I used both HD Tach (2.60) and ZD Labs Winbench 99 (ver. 1.1) disk test. Both test differently and can be used to judge the others scores for discrepancies. For ‘Real World’ performance I used ZD Labs Winstone 2001, both Business and Content Creation. I picked Winstone for a couple of reasons. I feel it reflects normal usage, since it actually installs the programs and uses scripts to run the program with real data. It also runs with a number of windows open at the same time, just like most of us do (I run with between 7 and 12 normally), and uses commonly used office and web tool programs. One other reason, I’ve been using it for years and have a pretty good ‘feel’ for whether it is giving the results it should or what I expect.
I used a Soyo 6BA+ 100 Intel 440BX based mainboard with the High Point Technology 370 ATA/100 RAID controller for these tests. When this mainboard was released (and it’s predecessor the 6BA+ IV with the non-RAID ATA/66 HPT 366 controller), I felt its best use and market was for those needing to connect more than 4 IDE devices. From my own testing I’d never seen any real performance advantage over the Intel BX built in ATA/33 controller and the ATA/66 or 100 HPT controllers, but I do see a number of users that had a need to connect more then four IDE devices. I didn’t feel there was much use (other than marketing) for the RAID function, and as a matter of fact, other than setting up a RAID array so I could offer support to customers I’d not even used it, much less done any benchmarks.
For the CPU, I decided to use a Pentium III 500E. After giving some thought to it I figured using a lower power CPU might keep pure CPU power from influencing the results. Just an idea. Remember that we want to isolate the disk performance as much as possible, though I could be way off base on this. For memory, I settled on 128MB, pretty much standard these days, and again, too little or too much could change the Winstone results due to swapping programs and data to and from the disk (more memory = less swapping, less = more) and the disk cache size changing.
Drives…I’m sure we could debate for days on the benefits and shortcomings of different brands and models of Hard Disk. I used two Western Digital 200BB 20GB ATA/100 drives. They are modern, up-to-date drives and what I’d consider upper end in performance. They should give a good indication of what the average user would see in drive performance. Besides, since that is what I sell the most of I have a couple of test units on hand, and availability is always an important issue. It also means I’m not going to be influenced or feel an obligation to someone if they were given to me.
I used Windows 98SE since it is the most common OS used. All settings were standard, with no ‘tweaking’ of the disk cache size or anything else. The latest BIOS from Soyo (2AA2) was used (it includes the HPT ver. 1.02 BIOS), along with the latest HPT ver. 1.03 driver. The BIOS was set for the best performance (CAS2, that sort of thing). An ATI Radeon DDR 32MB AGP card along with a D-Link 10/100 NIC were installed as was a 3.5 inch FD drive and an older PIO mode 4 ATAPI CDROM on the internal IDE port.
The disks were both partitioned with one 8GB partition each. This was to keep the data all in one area for more consistent test results. Results can vary if on one drive you are using the outer edge and the center on another.
The whole point was not to setup a system that would be the fastest but one that would give meaningful results based on an average quality system. These tests are to reflect usage for a stand-alone computer, and no effort has been made to compare or test under a server environment.
Be the first to discuss this article!