delayed disk block allocation
It's not on a big box, but I have a side by side of 2.5.6-pre2 and 2.5.6-pre2 with patches from
http://www.zipworld.com.au/~akpm/linux/patches/2.5/2.5.6-pre2/
2.5.6-pre2-akpm compiled with MPIO_DEBUG = 0 and ext2 filesystem mounted with delalloc.
tiobench and dbench were on ext2.
bonnie++ and most other tests were on reiserfs.
2.5.6-pre2-akpm throughput on ext2 is much improved.
http://home.earthlink.net/~rwhron/kernel/akpm.html
One odd number in lmbench is page fault latency. Lmbench also showed high page fault latency in 2.4.18-pre9 with make_request, read-latency2, and low-latency. 2.4.19-pre1aa1 with read_latency2 (2.4.19pre1aa1rl) did not show a bump in page fault latency.
Signature · 1 line
Randy Hron
Quoted 9 lines · [email protected] wrote:
2.5.6-pre2-akpm compiled with
MPIO_DEBUG= 0 and ext2 filesystem mounted with delalloc.tiobench and dbench were on ext2.
bonnie++ and most other tests were on reiserfs.2.5.6-pre2-akpm throughput on ext2 is much improved.
This page of statistics is pretty gnifty -- I wish more people posted such :)
I also like your other comparisons at the top of that link... I was thinking, what would be even more neat and more useful as well would be to post comparisons -between- the various kernels you have tested, i.e. a single page that includes benchmarks for 2.4.x-official, 2.5.x-official, -aa, -rmap, etc.
Regards,
Jeff
Signature · 3 lines
Jeff Garzik | Building 1024 | MandrakeSoft | Choose life.