[kernel-br] [PATCH] task_io_accounting, taskstats
Segue anexo o pequeno patch pra net taskstats accounting. Exemplo de userland: [root][hm2306][.c0] # cat ustatsd.log | grep -i comlines1 { "host" : "hm2306", "uid" : "11606", "name" : "comlines1", "time" : "2010:05:12:13:44", "ucpu" : "29864000", "pucpu" : "49.77", "scpu" : "5936000", "pscpu" : "9.89", "mem" : "1727.23" } { "host" : "hm2306", "uid" : "11606", "name" : "comlines1", "time" : "2010:05:12:13:45", "ucpu" : "30180000", "pucpu" : "50.29", "scpu" : "5236000", "pscpu" : "8.73", "mem" : "1535.17" } { "host" : "hm2306", "uid" : "11606", "name" : "comlines1", "time" : "2010:05:12:13:46", "ucpu" : "33064000", "pucpu" : "55.11", "scpu" : "5892000", "pscpu" : "9.82", "mem" : "1669.16" } O PUCPU eh porcentagem de utilizacao (ncpus * 100) e o PSCPU eh porcentagem de utilizacao de system do mesmo uid. Memoria hoje em dia eh um high watermark de todos procs que estao ativos e/ou morreram no timequantum medido. O userland gera um xmpp em json de mais de 1500 usuários por máquinas falando o consumo total de cpu, memoria, disco e (com meu patch) rede. Tem um frontend que gera os graficos de utilizacao de recursos por usuário. Abs Inaddy [root][macbook][linux-2.6.33.5]# patch -p1 < /home/inaddy/codes/ustats/us_2.6.33.x.patch patching file include/linux/task_io_accounting.h patching file include/linux/task_io_accounting_ops.h patching file include/linux/taskstats.h patching file net/socket.c CONFIG_TASKSTATS=y CONFIG_TASK_DELAY_ACCT=y CONFIG_TASK_XACCT=y CONFIG_TASK_IO_ACCOUNTING=y CONFIG_CONNECTOR=y CONFIG_PROC_EVENTS=y ---------- Forwarded message ---------- From: Rafael Tinoco <tinhocas@gmail.com> Date: Thu, Jun 3, 2010 at 10:07 PM Subject: Re: [PATCH] task_io_accounting, taskstats To: LKML <linux-kernel@vger.kernel.org>, Balbir <balbir@linux.vnet.ibm.com> Balbir, sorry Ive made a mistake sending the patch, it was an incomplete and wrong diff. And Ive fixed all the "tab" and signed-off-by issues. [inaddy][macbook][linux-2.6.33.5]$ ./scripts/checkpatch.pl /home/inaddy/codes/ustats/us_2.6.33.x.patch total: 0 errors, 0 warnings, 125 lines checked Tests to confirm patch is working, commands: # netcat -l -p 8080 # cat /etc/passwd | netcat localhost 8080 # cat /etc/passwd | wc -c 1126 # ./ustatsd (getting the 1126 bytes read and written for the netcat cmds) proc_event_exit: pid 3704 uid 0 (ucpu: 4000) (scpu: 0) (rdisk: 0) (wdisk: 0) (rnet: 0) (wnet: 1126) (comm: netcat) proc_event_exit: pid 3702 uid 0 (ucpu: 4000) (scpu: 0) (rdisk: 0) (wdisk: 0) (rnet: 1126) (wnet: 0) (comm: netcat) The attached patch is working. Ill be releasing in this same mail thread the userland code. I'll send another email with [PATCH] to the maintainer and to the list. Thank you. Rafael Tinoco
Olá Em 19-08-2010 10:46, tinhocas@gmail.com escreveu:
Segue anexo o pequeno patch pra net taskstats accounting.
Exemplo de userland:
[root][hm2306][.c0] # cat ustatsd.log | grep -i comlines1
{ "host" : "hm2306", "uid" : "11606", "name" : "comlines1", "time" : "2010:05:12:13:44", "ucpu" : "29864000", "pucpu" : "49.77", "scpu" : "5936000", "pscpu" : "9.89", "mem" : "1727.23" } { "host" : "hm2306", "uid" : "11606", "name" : "comlines1", "time" : "2010:05:12:13:45", "ucpu" : "30180000", "pucpu" : "50.29", "scpu" : "5236000", "pscpu" : "8.73", "mem" : "1535.17" } { "host" : "hm2306", "uid" : "11606", "name" : "comlines1", "time" : "2010:05:12:13:46", "ucpu" : "33064000", "pucpu" : "55.11", "scpu" : "5892000", "pscpu" : "9.82", "mem" : "1669.16" }
O PUCPU eh porcentagem de utilizacao (ncpus * 100) e o PSCPU eh porcentagem de utilizacao de system do mesmo uid. Memoria hoje em dia eh um high watermark de todos procs que estao ativos e/ou morreram no timequantum medido.
Bacana, mais uma medida. Mas deixa eu entender, tu precisa de ajuda pra montar uma justificativa para os patches/new features? isso? Abraço... -- Dorileo Atenciosamente.... -- (°= Leandro Dorileo //\ ldorileo@gmail.com - http://www.dorilex.net V_/ Software is a matter of freedom.
participants (2)
-
Leandro Dorileo -
tinhocas@gmail.com