Tek-Tips is the largest IT community on the Internet today!

Members share and learn making Tek-Tips Forums the best source of peer-reviewed technical information on the Internet!

  • Congratulations wOOdy-Soft on being selected by the Tek-Tips community for having the most helpful posts in the forums last week. Way to Go!

CPU perf measurement: top vs. vmstat vs. mpstat

Status
Not open for further replies.

pcorchary

MIS
Feb 23, 2002
60
US
[red]I am looking for feedback from experienced Linux admins regarding CPU usage analysis in Linux. The following should be taken in context of dual-CPU (Intel) systems with RedHat 7.2, kernel 2.4.7.10smp.[/red]

1) Has anyone seen issues with 'top' (from the procps package), where when the system has been up for more than about 200 days, top gives incorrect information and in fact core dumps when run!? (top from procps 2.07 -- yes, there are newer versions, but there glibc dependencies that prevent us from upgrading at this time.)

2) how far to you trust/believe top vs. vmstat, especially since that vmstat aggregates CPU information from all CPUs in an SMP system (correct?).

3) how useful is mpstat (from the sysstat package)? It's not part of our production environment now, and I would have to make the case to add it to hundreds of systems :)

Background is that I'm still somewhat new to Linux (but not UNIX), and we're seeing problems with top output, but I'm recommending the use of vmstat and mpstat based on my prev. experience in HP and Solaris.

Thanks in advance!
 

1) I have never seen a Linux system that was up for more than 200 days, so I wouldn't know.

2) I think the difference is what is displayed. vmstat regards the system as one processing unit where top adds up each individual processor (I think). vmstat is probably the standard way of doing it.

In my experience top gives reliable output you just have to know how to interpret it. What exactly is the problem??

Cheers Henrik Morsing
Certified AIX 4.3 Systems Administration
& p690 Technical Support
 
Well, I have several hundred Linux boxes, many over 200, some over 300, and a few over 400 days uptime :)

the problem is that in the described configuration, after about 200 days uptime, we see that top gives VERY incorrect information, and in many cases core dumps immediately after invocation!

In the Sun and HP environments I long ago learned not to trust top ...

philc
 
Status
Not open for further replies.

Part and Inventory Search

Sponsor

Back
Top