Bad Performance CentOS6 ->CentOS7
hi,
i’m evaluating performance of CentOS7
i did tests on CentOS6 x86_64/distro kernel 2.6.32, asterisk 11.16.0
with 500calls (7sec alaw, simple dialplan, pass trough – sipp generators/asterisk receiver with answer/playback)
scenario – sipp generators – asterisk – asterisk receiver (i wrote ansible scenario for this if you are interested)
then i reinstalled system to CentOS7 x86_64/distro kernel 3.10, asterisk 11.20.0 and run the test again
there is big performance hit https://dl.dropboxusercontent.com/u/44105720/context_interrupts.PNG
https://dl.dropboxusercontent.com/u/44105720/cpu.PNG
https://dl.dropboxusercontent.com/u/44105720/load.PNG
any ideas what tweaks can help? (it looks like the main problem is in interrupts from network card)
your experience with CentOS7?
any experience with kernel 4.2 from http://elrepo.org/linux/kernel/el7/x86_64/RPMS/ ?
2 thoughts on - Bad Performance CentOS6 ->CentOS7
found this interesting article http://stackoverflow.com/questions/12111954/context-switches-much-slower-in-new-linux-kernels
running with network-latency profile and its better now but still not good as CentOS6
[root@ast1 ~]# tuned-adm active Current active profile: network-latency
Dne 6.11.2015 v 10:18 Marek Červenka napsal(a):
—
—————————————
Marek Cervenka
=======================================
—
found interesting difference
CentOS6 have increasing “Function call interrupts” only on cpu4 (its 2x quad core system with HT cpu0-cpu15)
RES: 6286605 2569745 1885919 1842514 2177536 2138628
1761718 1838163 3668947 2597374 1849095 1679889
2736591 2523227 1751432 1611744 Rescheduling interrupts
CAL: 248 362 366 362 40923894
349 357 361 345 305 364 367
363 337 356 352 Function call interrupts
CentOS7 have “Function call interrupts” distributed to all cpus RES: 18659 22826 17808 14357 42245 36973
29927 24703 15279 9098 7984 6564 27054
23757 19585 16525 Rescheduling interrupts
CAL: 304288 30969 30869 30555 39649 34481
32515 28972 11286 22461 22623 22792 25012
24027 22610 21718 Function call interrupts
any ideas?
Dne 9.11.2015 v 13:28 Marek Červenka napsal(a):
—
—————————————
Marek Cervenka
=======================================
—