I sent this information to Andrea while he was tuning his elevator patch,
but since this has come up on the list, I thought others might have some
use for the testing I did.
Basically I tried various settings for the ELEVATOR_DEFAULT in attempt to
boost the fall in the rewrite figure generated from bonnie.
My machine, btw: UP P2/400, 128meg IDE.
Here's my results, including a pre-elevator kernel for reference:
-------Sequential Output-------- ---Sequential Input-- --Random--
-Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU
2.3.42
500 6061 98.8 11995 9.2 4291 5.5 5527 82.7 11322 8.3 89.0 1.5
2.4.46
vanila: { 0, NR_REQUEST>>1, NR_REQUEST<<5, 4, 0, })
500 6013 98.4 9097 7.1 2686 3.4 5806 86.8 11155 8.0 83.6 1.4
500 6034 98.5 9116 6.7 2692 3.4 5751 86.0 11155 7.8 83.9 1.5
a: { 0, NR_REQUEST>>1, NR_REQUEST<<5, NR_REQUEST>>2, 0, })
500 6046 98.9 8462 6.4 2660 3.5 5715 85.3 10907 7.6 86.0 1.2
b: { 0, NR_REQUEST, NR_REQUEST<<5, 4, 0, })
500 6007 98.1 8462 6.5 2677 3.6 5787 86.6 11044 7.9 85.1 1.3
c: { 0, NR_REQUEST<<2, NR_REQUEST<<5, 4, 0, })
500 5925 96.5 8487 6.5 3042 3.6 5762 86.1 11041 7.8 86.4 1.6
Obviously everyone knows that its just a dumb benchmark...
The a, b, and c tests had different ELEVATOR_DEFAULTS as shown.
c. kernel had a improved rewrite figure, but it made the machine feel
alittle strange. Hard to say exactly how, but it just didn't feel quite
as silky smooth as the vanilla 2.3.46. Would this dbench test be impeeded
by the fall in rewrite figures? Andrea, you had a fix for the 2.2 rewrite
drop. I supose it can't be made to work on 2.3 or you would have done it?
Incidently Ive been using 2.3 now for about a month and it works great! I
haven't had a single problem or hickup, even through my own stress tests
of a load of 50 (50 simultanious compiles, heavy swapping etc) and still
this machine is useable, and from the console the load is easily
acceptable.
I am sorry if I stepped on your toes, Andrea, by posting the figures. :((
Lawrence
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Tue Feb 29 2000 - 21:00:17 EST