+46% netperf.Throughput_Mbps
Subsystems (2)
Hi Eric,
Just FYI, we do measured +46.4% performance boost in netperf TCP_SENDFILE tests on an atom server for commit:
commit 0cf833aefaa85bbfce3ff70485e5534e09254773
Author: Eric Dumazet <[email protected]> AuthorDate: Sun Sep 23 22:28:59 2012 +0000 Commit: David S. Miller <[email protected]> CommitDate: Mon Sep 24 16:23:53 2012 -0400
net: loopback: set default mtu to 64K
loopback current mtu of 16436 bytes allows no more than 3 MSS TCP segments per frame, or 48 Kbytes. Changing mtu to 64K allows TCP stack to build large frames and significantly reduces stack overhead.
Performance boost on bulk TCP transferts can be up to 30 %, partly because we now have one ACK message for two 64KB segments, and a lower probability of hitting /proc/sys/net/ipv4/tcp_reordering default limit.
drivers/net/loopback.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
2a6c8c7998f95b1 0cf833aefaa85bbfce3ff7048
--------------- -------------------------
1233 ~ 4% +46.4% 1805 ~ 0% netperf.Throughput_Mbps
29214 ~ 9% -85.2% 4310 ~17% interrupts.RES
450970 ~ 3% -57.3% 192675 ~ 1% proc-vmstat.pgalloc_normal
453096 ~ 3% -57.0% 194790 ~ 1% proc-vmstat.pgfree
4664799 ~ 6% -47.7% 2441398 ~ 0% softirqs.NET_RX
1107 ~ 9% +62.2% 1797 ~ 0% slabinfo.kmalloc-1024.active_objs
1185 ~ 8% +54.2% 1827 ~ 0% slabinfo.kmalloc-1024.num_objs
244825 ~ 3% -28.6% 174791 ~ 0% proc-vmstat.numa_hit
244825 ~ 3% -28.6% 174791 ~ 0% proc-vmstat.numa_local
913 ~ 2% +39.3% 1271 ~ 1% slabinfo.skbuff_fclone_cache.active_objs
944 ~ 3% +36.9% 1293 ~ 1% slabinfo.skbuff_fclone_cache.num_objs
10 ~ 0% -10.0% 9 ~ 0% vmstat.procs.r
59630 ~12% -94.4% 3336 ~ 8% vmstat.system.cs
1233 ~ 1% -16.4% 1031 ~ 0% vmstat.system.innetperf.Throughput_Mbps
2000 ++------------------------------------------------------------------+ 1800 O+O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O O
| |
1600 ++ |
1400 ++ *. |
*.*.*.. .*.*. + * *.*.*.*. .*.*. .*.* |
1200 ++ *.* *..* : : *.*. * |
1000 ++ : : |
800 ++ : : |
| : : |
600 ++ : : |
400 ++ : : |
| : : |
200 ++ : : |
0 ++----------------------*-*-----------------------------------------+