[Orca-users] collecting process information

[Glen Gunselman]

I've considered adding something like that to Big Brother.  
 
Does extended accounting collect enough detail to fill in those blanks ...
 
Glen Gunselman
Systems Software Specialist
TCS
Emporia State University

>>> David Michaels <dragon at raytheon.com> 03/14/07 5:01 PM >>>
It sounds like what Joe is looking for is information on each individual process at a given point in time.  For instance, if the load-average graph shows a sharp spike in CPU usage, Joe wants Orca to tell him which process is causing the spike.  At least, that's what I'm getting from the question.

Joe -- I don't think this is something you can do with orca with the out-of-the-box collectors for Solaris.  You could create a script that generated the data you wanted, then just include that data source in your orcallator.cfg file.  I haven't tried something like that on the Sun side, so it may be more involved than that.

I imagine what you'd want to do is track the CPU usage of every process, by PID.  You will end up with an enormous amount of data, so I personally would not do it for long-term tracking.  Perhaps limiting it to the top 5-10 PIDs by cpu time & memory usage or something would help to reduce the data substantially while keeping it valuable.

--Dragon

Glen Gunselman wrote: 


Joseph,
 
What details are you looking for?
 
The Orca collection file is somewhat self describing.  If you look at the first record in the file you will see the fields it contains.  Here what one of mine looks like:
 
timestamp  locltime   uptime state_D state_N state_n state_s state_r state_k state_c state_m state_d state_i state_t DNnsrkcmdit  us
r%  sys%  wio% idle%  1runq  5runq 15runq #proc  #runque #waiting  #swpque scanrate #proc/s #proc/p5s  smtx smtx/cpu ncpus mntC_/ mn
tU_/ mntA_/ mntP_/ mntc_/ mntu_/ mnta_/ mntp_/ mntC_/var mntU_/var mntA_/var mntP_/var mntc_/var mntu_/var mnta_/var mntp_/var mntC_
/export mntU_/export mntA_/export mntP_/export mntc_/export mntu_/export mnta_/export mntp_/export mntC_/opt/openv mntU_/opt/openv m
ntA_/opt/openv mntP_/opt/openv mntc_/opt/openv mntu_/opt/openv mnta_/opt/openv mntp_/opt/openv mntC_/nbu/disk mntU_/nbu/disk mntA_/n
bu/disk mntP_/nbu/disk mntc_/nbu/disk mntu_/nbu/disk mnta_/nbu/disk mntp_/nbu/disk mntC_/nbu/staging mntU_/nbu/staging mntA_/nbu/sta
ging mntP_/nbu/staging mntc_/nbu/staging mntu_/nbu/staging mnta_/nbu/staging mntp_/nbu/staging disk_runp_c0t0d0 disk_runp_c1t0d0 dis
k_runp_c1t1d0 disk_runp_c1t2d0 disk_runp_c1t3d0 disk_runp_md10 disk_runp_md20 disk_runp_md0 disk_runp_md13 disk_runp_md23 disk_runp_
md3 disk_runp_c5t600A0B800017741C000025B44289FDCEd0 disk_runp_c5t600A0B8000176F750000149D4289FC45d0 disk_runp_md11 disk_runp_md21 di
sk_runp_md1 disk_runp_md50 disk_runp_md51 disk_runp_md5 disk_runp_md100 disk_runp_md101 disk_runp_md102 disk_peak disk_mean disk_rd/
s disk_wr/s disk_rK/s disk_wK/s swap_avail page_rstim   freememK free_pages   ce0Ipkt/s   ce0Opkt/s   ce0InKB/s   ce0OuKB/s   ce0IEr
r/s   ce0OErr/s   ce0Coll%   ce0NoCP/s   ce0Defr/s   ce1Ipkt/s   ce1Opkt/s   ce1InKB/s   ce1OuKB/s   ce1IErr/s   ce1OErr/s   ce1Coll
%   ce1NoCP/s   ce1Defr/s tcp_Iseg/s tcp_Oseg/s tcp_InKB/s tcp_OuKB/s tcp_Ret% tcp_Dup% tcp_Icn/s tcp_Ocn/s tcp_estb tcp_Rst/s tcp_A
tf/s tcp_Ldrp/s tcp_LdQ0/s tcp_HOdp/s nfs_call/s nfs_timo/s nfs_badx/s nfss_calls nfss_bad  v2reads v2writes  v3reads v3writes dnlc_
ref/s dnlc_hit% inod_ref/s inod_hit% inod_stl/s pp_kernel pagesfree pageslock pagestotl
 
Glen Gunselman
Systems Software Specialist
TCS
Emporia State University

>>> "Beck, Joseph" <jbeck at seic.com> ( mailto:jbeck at seic.com ) 03/14/07 2:10 PM >>>

We currently use orca throughout our environment(120 sun boxes) & it's useful for many reasons. 
One glaring piece missing, though, is process information.
I haven't setup orca in the past & in this environment I've only added agents/clients.
So, I'm not sure if this process info gap is specific to our implementation or if it's not there in orcaby default.
 
Either way, my goal is to collect process info in order to be able to connect the dots between a load (or spike) & a process.
Is this capability there out of the box with orca? If not, has there been any efforts to leverage prstat, psio.se, etc. to collect & graph it?
Ultimately, I'd like to get to organizing processes into workloads & begin to understand our application resource utilization.
 
Thanks for any feedback or references,
 
Joe Beck Ciber Inc. - a consultant to SEI One Freedom Valley Drive/100 Cider Mill Road| Oaks, PA 19456 | p: 610.676.2258 | jbeck at seic.com
 
_______________________________________________
Orca-users mailing listOrca-users at orcaware.comhttp://www.orcaware.com/mailman/listinfo/orca-users  
-------------- next part --------------
An HTML attachment was scrubbed...
URL: </pipermail/orca-users/attachments/20070314/5d720675/attachment-0002.html>