Winbench 99 on different partitions UDMA 100 IBM
Have just upgraded my PC with a PCI UDMA 100 CMD Controller and an IBM DTLA 307045 hard disk (46 GB, 7200 rpm, UDMA 100). Everything runs OK on my W98SE OS. Wanted to test the performance with the new system running the Winbench 99 disk test. My disk was partitioned with FAT32 using FDISK, have in total 8 partitions 6,8; 7,8; 7,8; 2,4; 2,4; 1,5; 2,4 and 11,4 GB. The first partition (6,8 GB) is the primary one on which my W98SE is installed. Other partitions are logical extended DOS partitions. Runing the benchmark test in UDMA 100 Mode 5 I am finding significant differencies in the transfer rate between different partitions as follows: C:\ 38 MB/s; D:\ 36; E:\ 33; F:\ 31; G:\ 29; H:\ 29; I:\ 27; J:\ 27. While runung the test in Mode 2 (UDMA33), there are almost no differencies in transfer rates between different partitions. I can change the UDMA Mode in my CMD 649 Windows drivers. Can somebody explain why? How can I get better performance on my extended partitions? Does it make sense to have several partitions if UDMA5 applies on the primary partition only?
What your seeing is normal.
Partitions at the outside of the disk have faster access/transfer then those at the inside of the disk. If you look at good HD reviews they list the performance at the beginning (outside) of the drive as well as the end (inside).
It's easy to explain. The data density is the same everywhere on the disk, but the outside of the disk moves under the heads faster then the inside does, thus the higher rates.
In theory, there is no difference between theory and practice.
In practice, there is.
What all this is saying to me is that the deeper towards the disk centre you go the more is the transfer rate controlled by the disk performance rather than by the UDMA mode used. Following this thought I am questioning the way of partitioning disks. For example if your primary partition is very big and you are using just 20-30% of its space, you are most probably "throwing away" some performance for your next closest partition. Correct? Has anybody done some studies in this direction?