[PATCH 1/1] namespaces: introduce sys_hijack (v11)
Serge E. Hallyn
serue at us.ibm.com
Thu Aug 7 12:23:38 PDT 2008
Quoting KOSAKI Motohiro (kosaki.motohiro at jp.fujitsu.com):
> Hi
>
> fork() is very important for system performance.
> I worry about performance regression if this feature isn't used.
>
> Could you mesure spawn benchmark in unixbench?
Sigh, well I did run some benchmarks on a machine on our grid, but none
seem to be giving useful or consistent results. Unixbench results are
attached in the following files:
unixbench results:
elm3b165.nohijack.1: result of './Run 1' on plain 2.6.27-rc1
elm3b165.nohijack.4: result of './Run 4' on plain 2.6.27-rc1
elm3b165.withhijack.1: result of './Run 1' on 2.6.27-rc1 with hijack patch
elm3b165.withhijack.4: result of './Run 4' on 2.6.27-rc1 with hijack patch
I suppose I could have automated these to get more runs, but as it is
the results seem meaningless to me.
I also ran dbench and tbench. All tbench instances were with
5 clients, all dbench with 4 clients. I ran both dbench and tbench 10
times on each kernel, and here are the mean throughput +/- 95% confidence
interval:
Plain 2.6.27-rc1 kernel:
tbench: 520.355900 +/- 17.104307
dbench: 671.337900 +/- 146.785891
2.6.27-rc1 with hijack:
tbench: 523.213300 +/- 19.726943
dbench: 437.324400 +/- 169.086289
The tbench ones are solidly within the confidence bounds. The dbench
results aren't really believable. Cron wasn't running on the machine,
and nothing else *should* have been, but alas I don't own the machine
so don't know what could have been running.
thanks,
-serge
-------------- next part --------------
INDEX VALUES
TEST BASELINE RESULT INDEX
Dhrystone 2 using register variables 376783.7 8167048.5 216.8
Double-Precision Whetstone 83.1 1199.0 144.3
Execl Throughput 188.3 6777.2 359.9
File Copy 1024 bufsize 2000 maxblocks 2672.0 113916.0 426.3
File Copy 256 bufsize 500 maxblocks 1077.0 36714.0 340.9
File Read 4096 bufsize 8000 maxblocks 15382.0 718796.0 467.3
Pipe Throughput 111814.6 1425864.8 127.5
Pipe-based Context Switching 15448.6 394297.5 255.2
Process Creation 569.3 18977.8 333.4
Shell Scripts (8 concurrent) 44.8 1217.8 271.8
System Call Overhead 114433.5 2858840.5 249.8
=========
FINAL SCORE 270.3
INDEX VALUES
TEST BASELINE RESULT INDEX
Dhrystone 2 using register variables 376783.7 8362581.0 221.9
Double-Precision Whetstone 83.1 1203.5 144.8
Execl Throughput 188.3 6715.0 356.6
File Copy 1024 bufsize 2000 maxblocks 2672.0 107020.0 400.5
File Copy 256 bufsize 500 maxblocks 1077.0 32403.0 300.9
File Read 4096 bufsize 8000 maxblocks 15382.0 678707.0 441.2
Pipe Throughput 111814.6 1385493.0 123.9
Pipe-based Context Switching 15448.6 372454.9 241.1
Process Creation 569.3 18607.9 326.9
Shell Scripts (8 concurrent) 44.8 1208.8 269.8
System Call Overhead 114433.5 2691682.8 235.2
=========
FINAL SCORE 260.7
-------------- next part --------------
INDEX VALUES
TEST BASELINE RESULT INDEX
Dhrystone 2 using register variables 376783.7 8232510.9 218.5
Double-Precision Whetstone 83.1 1202.4 144.7
Execl Throughput 188.3 6739.0 357.9
File Copy 1024 bufsize 2000 maxblocks 2672.0 104621.0 391.5
File Copy 256 bufsize 500 maxblocks 1077.0 33092.0 307.3
File Read 4096 bufsize 8000 maxblocks 15382.0 684132.0 444.8
Pipe Throughput 111814.6 1344322.6 120.2
Pipe-based Context Switching 15448.6 385326.1 249.4
Process Creation 569.3 18701.0 328.5
Shell Scripts (8 concurrent) 44.8 1211.8 270.5
System Call Overhead 114433.5 2910270.2 254.3
=========
FINAL SCORE 262.6
INDEX VALUES
TEST BASELINE RESULT INDEX
Dhrystone 2 using register variables 376783.7 8247240.2 218.9
Double-Precision Whetstone 83.1 1198.4 144.2
Execl Throughput 188.3 6540.9 347.4
File Copy 1024 bufsize 2000 maxblocks 2672.0 97453.0 364.7
File Copy 256 bufsize 500 maxblocks 1077.0 31134.0 289.1
File Read 4096 bufsize 8000 maxblocks 15382.0 781214.0 507.9
Pipe Throughput 111814.6 1408780.0 126.0
Pipe-based Context Switching 15448.6 351906.0 227.8
Process Creation 569.3 18638.3 327.4
Shell Scripts (8 concurrent) 44.8 1211.8 270.5
System Call Overhead 114433.5 2737286.8 239.2
=========
FINAL SCORE 259.3
-------------- next part --------------
INDEX VALUES
TEST BASELINE RESULT INDEX
Dhrystone 2 using register variables 376783.7 8899355.0 236.2
Double-Precision Whetstone 83.1 1204.8 145.0
Execl Throughput 188.3 6626.6 351.9
File Copy 1024 bufsize 2000 maxblocks 2672.0 109426.0 409.5
File Copy 256 bufsize 500 maxblocks 1077.0 34428.0 319.7
File Read 4096 bufsize 8000 maxblocks 15382.0 625814.0 406.8
Pipe Throughput 111814.6 1296658.5 116.0
Pipe-based Context Switching 15448.6 359970.6 233.0
Process Creation 569.3 18747.0 329.3
Shell Scripts (8 concurrent) 44.8 1210.8 270.3
System Call Overhead 114433.5 2745197.3 239.9
=========
FINAL SCORE 260.2
INDEX VALUES
TEST BASELINE RESULT INDEX
Dhrystone 2 using register variables 376783.7 8823465.9 234.2
Double-Precision Whetstone 83.1 1205.6 145.1
Execl Throughput 188.3 6603.0 350.7
File Copy 1024 bufsize 2000 maxblocks 2672.0 104783.0 392.2
File Copy 256 bufsize 500 maxblocks 1077.0 32153.0 298.5
File Read 4096 bufsize 8000 maxblocks 15382.0 859178.0 558.6
Pipe Throughput 111814.6 1185154.3 106.0
Pipe-based Context Switching 15448.6 344165.8 222.8
Process Creation 569.3 18397.4 323.2
Shell Scripts (8 concurrent) 44.8 1207.6 269.6
System Call Overhead 114433.5 2654906.5 232.0
=========
FINAL SCORE 260.3
-------------- next part --------------
INDEX VALUES
TEST BASELINE RESULT INDEX
Dhrystone 2 using register variables 376783.7 8844791.9 234.7
Double-Precision Whetstone 83.1 1206.1 145.1
Execl Throughput 188.3 6700.7 355.9
File Copy 1024 bufsize 2000 maxblocks 2672.0 102895.0 385.1
File Copy 256 bufsize 500 maxblocks 1077.0 30585.0 284.0
File Read 4096 bufsize 8000 maxblocks 15382.0 877998.0 570.8
Pipe Throughput 111814.6 1146617.1 102.5
Pipe-based Context Switching 15448.6 355382.2 230.0
Process Creation 569.3 18548.7 325.8
Shell Scripts (8 concurrent) 44.8 1208.8 269.8
System Call Overhead 114433.5 2815676.2 246.1
=========
FINAL SCORE 261.2
INDEX VALUES
TEST BASELINE RESULT INDEX
Dhrystone 2 using register variables 376783.7 8884179.7 235.8
Double-Precision Whetstone 83.1 1202.7 144.7
Execl Throughput 188.3 6543.5 347.5
File Copy 1024 bufsize 2000 maxblocks 2672.0 103051.0 385.7
File Copy 256 bufsize 500 maxblocks 1077.0 32475.0 301.5
File Read 4096 bufsize 8000 maxblocks 15382.0 668682.0 434.7
Pipe Throughput 111814.6 1269063.8 113.5
Pipe-based Context Switching 15448.6 357748.7 231.6
Process Creation 569.3 18862.8 331.3
Shell Scripts (8 concurrent) 44.8 1206.8 269.4
System Call Overhead 114433.5 2754854.2 240.7
=========
FINAL SCORE 258.1
More information about the Containers
mailing list