performed some tests days before and I would like to share the results with all of you. the result was as expected (my expectation)

the test was conducted on /dbdata, using dd to read and write, and with different slice_sync and slice_async values.
According to lwn.net:
- slice_sync: How many msec a sync disk slice lasts
- slice_async: How many msec an async disk slice lasts

in froyo, the default value for slice_sync is 97, and slice_async is 39. while in eclair, they are 100 and 40 respectively.

and, the results were obtained by the average value under different file systems as below:
write: dd if=/dev/zero of=/dbdata/test count=25000 (using default bs=512)
read: dd if=/dbdata/test of=/dev/null

test1: default slice_sync (97), slice_async (39)
ext2 w=0.3563576, r=0.2172932
ext4 w=1.7057048, r=0.1286368
ext2 on ext4 w=0.2786272, r=0.1360666
ext2 on ext4 noatime nodiratime w=0.279215, r=0.124138
ext4 on ext2 w=0.4299866, r=0.1277804

test2: slice_sync (50), slice_async (20)
ext2 w=0.3883144,r=0.2209398
ext4 - omitted
ext2 on ext4 w=0.2743988,r=0.1343098
ext4 on ext2 w=0.4350612,r=0.2513572

test3: slice_sync (500), slice_async (200)
ext2 w=0.4159796, r=0.40419
ext4 - omitted
ext2 on ext4 - omitted
ext4 on ext2 w=0.4252074, r=0.2614818

obviously, the fastest one was ext2 on top of ext4, with only insignificant impact with noatime and nodiratime options (i cant believe it). this combination of file systems performed well as expected since while ext2 do somewhat like "blind read/write", the ext4 will hold the data b4 commit (PLS, pls dont argue ext2 and ext4 with me and that's why i described them very roughly... )


actually this was just for my own leasure but i think it may be useful to u guys as well so i decided to post it here for your ref


more info