Re: How to troubleshoot this without dtrace?
- From: andrew@xxxxxxxxxxxxxxxxxxxx (Andrew Gabriel)
- Date: 03 Jun 2008 15:30:20 GMT
In article <b68d4256-b0de-4f94-986e-a48229e8f160@xxxxxxxxxxxxxxxxxxxxxxxxxxxx>,
Jim Leonard <MobyGamer@xxxxxxxxx> writes:
(and I'd like the answer to be something other than "upgrade to
sol10" ;-)
I have a solaris 8 machine with sar output that looks like this:
00:00:01 %usr %sys %wio %idle
08:20:00 14 10 2 74
08:40:00 12 8 2 79
09:00:00 16 49 1 34
09:20:00 16 56 1 27
09:40:00 17 47 1 35
You can see at the 9:00:00 mark that the %sys column is starting to
climb, although prstat doesn't show anything new than previously (ie.
all of the same daemons are running, no new user procs).
How do I start investigating this? Is there a sar option (or
something else) that is especially suited to determining what the OS
is spending its time on?
Are you only able to look at the historical sar data, or does this
happen every 09:00:00 so you can look again tomorrow and collect
additional data?
I would start with things like vmstat, iostat, prstat -m,
intrstat. It's quite easy (and common) for a process to cause
significant %sys without actually using any significant %usr.
Is the system an NFS server? Could be NFS traffic.
If you only have that historical data, look at sar -A output.
Even on a system with dtrace, that wouldn't be my first tool
for this problem as described.
--
Andrew Gabriel
[email address is not usable -- followup in the newsgroup]
.
- Follow-Ups:
- Re: How to troubleshoot this without dtrace?
- From: Jim Leonard
- Re: How to troubleshoot this without dtrace?
- References:
- How to troubleshoot this without dtrace?
- From: Jim Leonard
- How to troubleshoot this without dtrace?
- Prev by Date: Re: Automatic installation of many X4500
- Next by Date: Re: Automatic installation of many X4500
- Previous by thread: Re: How to troubleshoot this without dtrace?
- Next by thread: Re: How to troubleshoot this without dtrace?
- Index(es):
Relevant Pages
|