I've got an apparently common problem, and I've been scouring newsgroups/discussion boards for an answer, but the threads are always left hanging--there is never an appropriate solution. So here's my detailed troubleshooting report of this problem as I've experienced it over many months. I hope it can inspire new way to fix this. Please read the whole thing before speculating, as it will take an entire reading before appreciating how I've tried to fix this problem already. My system specs are at the bottom of this post.
Problem in a metaphor: There is a bottleneck in one of my boxes. It is obvious to any user who uses many machines as it manifests, among other ways, by a delay between clicking an icon and having that application run. This delay is 12+ seconds and is not too dependent on the application size or the hard drive on which the application lies. It happens even with windows explorer and internet explorer. SISOFT Sandra pointed this out at one time as a slow hard drive or something like that, and pcpitstop noted it more recently. Pcpitstop was more specific in it's diagnosis: Drive C has a cached speed of 26.13 megabytes per second. Drive C has an uncached speed of 1.41 megabytes per second. Drive D has a cached speed of 25.49 megabytes per second. For comparison, systems with the same CPU, clock speed, and memory size as this one have an average cached speed of 62.51 MB/s and an uncached speed of 2.22 MB/s.
Obvious suggestion 1: hard drive is too old/small/slow.
Reply: This was my thought as well. I upgraded to a WD Caviar 15.3GB EIDE UDMA/66.6MB/s 9.0ms 7200RPM using an UDMA66 PCI card so the mainboard could handle its speed. And it can. Settings, drivers, and diagnostics confirm the high transfer rate from the drive. That being said, this system now exists with 2 hard drives. The above one is now D, and C is a Samsung 6.4GB EIDE UDMA/33 9.5MS 5400RPM. Trying both drives together and singly did not fix the bottleneck. In case the reader (you) didn't read that the disks are operating at the correct speeds, I am aware that in order for a UDMA to use it's capabilities it cannot be on the same cable with any non-UDMA device. The devices are not on the same cable. I've got 4 IDE headers now with the UMDA card, so no devices have to share a cable.
Obvious suggestion 2: RAM is limited by size/speed/mixing SDRAM modules.
Reply: I tried this. The SDRAM is now PC100, ECC, 128MB, 8ns, CL2 in a single module. I've turned ECC on and off. This does not change the bottleneck. The RAM is in Bank 0 of 3 on the mainboard. Regarding RAM size, the mainboard manual writes in one part supports 64M-bits (8Mx8, 4Mx16) EDO DRAM/SDRAM while in another part it says it supports 384MB (128 x 3 banks). I thought perhaps I had to use 64MB modules max per bank because of the first reference, but the second reference shows 128MB per bank is okay, so I don't think that is the problem.
Obvious suggestion 3: hard drive needs defragging and/or other maintenance.
Reply: I run the Maintenance Wizard and, independently of it, defrag often. After defragging, there was no change in the bottleneck as measured by pcpitstop.
Suggestion 4: cache requires management with Cacheman et al.
Reply: I tried the cache manipulations in Cacheman, Winrambooster PRO, and TweakME Gold (each separately) with no significant changes as measured by pcpitstop.
Suggestion 5: turn off the external cache in the bios.
Reply: I tried various combinations of external cache on/off and the system bios cached on/off with no change in the bottleneck as measured by pcpitstop.
My speculations: something is wrong with the mainboard or the OS management of resources. Maybe there is some commonality among users with this problem like VIA chipsets or PCCHIPS mainboards or RAM size/brand/speed or WinME etc. One post I read from someone quite knowledgeable with the cache wrote, "cached performance figures [are] related to RAM access speed and to CPU data bus width." Both of these areas are limited compared to more modern CPUs & RAM modules but not nearly to the extent that would produce the results reported by pcpitstop. And SANDRA benchmarks the C drive at Index 3700, a level comparable to a plain old ATA 4GB drive (it should be Index 8000).
Aside from the bottleneck, the system works great. It's totally stable (even more stable than my newer boxes running win2k!) and it's always on. Others experiencing this problem have no other system problems as well, as I have learned by reading many posts. Because everything else works great, fixing this is not critical. But wouldn't it be nice if I could get data moving as fast and with as much throughput as my components should permit? I think so. As far as I can tell, there has never been a fix put forth that works.
System specs: AMD K6-2 450mhz, on a PCCHIPS m577 mobo with VIA Apollo MVP3 chipset with latest VIA4in1 drivers (here are some known problems with the 577 http://www.geocities.com/SiliconValley/Peaks/9026/m577stuf.htm ; user reviews are here http://www.sysopt.com/userreviews/mboards/reviewhtml/PC_Chips_M577.html ), network card, Voodoo5 AGP, ISA soundcard, USB devices, running Windows ME.
Please cc: responses to 4chaos@altavista.com if you can as I am posting this to multiple discussion lists. Thus it may be hard for me to follow up on all lists unless you cc: to me.
Thanks!
