Eric Auer writes:
Hi, I have done some harddisk speed benchmarks on my system. I found UDMA to work well (but it is incompatible to some of the DOS extenders used by some programs, at least for me - anybody else?). I found LBAcache to work well, too, but ONLY if RAM is fast enough. For the normal RAWREAD tests, an access pattern is used which can hardly be cached, so those modes show how much OVERHEAD the cache processing is causing. The "cached" test modes of RAWREAD show how fast data is transferred for the cases where the cache can help.
I would like to see more results:
- for more "normal" access patterns (i.e. neither RAWREAD normal nor RAWREAD "cached" test modes)
- for different memory managers (possibly relevant: A20 / HMA, XMS and UMB speed, as well as load location (low <-> UMB) of drivers)
- for different cache sizes
- (invent other benchmarks yourself)
The bad news: If your RAM is slow enough and your disk is fast enough, you better work without LBAcache, or only with a SMALL LBAcache. However, a really BIG LBAcache will allow the disk to stay idle (and even spin down if you activated that in BIOS) most of the time. Slow but energy- saving and quiet operation. In either case, loading UDMA is a good idea unless you have compatibility problems.
Note: comparison tables and discussion of results can be found at http://www.coli.uni-sb.de/~eric/stuff/soft/specials/disk-speed-test-dos.txt