Appaling performance on Solaris 10 running on M5000 Server

[kaushikbramesh]

Hi

We have Solaris 10 running on a M5000 server.
We are trying to investigate a serious performance issue on the server. We have looked at various things such as I/O, CPU, memory, network...

CPU
The CPU seems to be 80% idle.

Disk I/O
isk I/O is not so bad.
We have occasional peak on /tmp

Memory utilization
Around 40% of memory is free, the system is not swapping too much.

Network
the network utilization is around 50%.

The system has 64 cores 128GB physical memory, 1GB network (fibre), 4 128GB internal disks. We have mirrored 2 disks together to make it two separate mirrored volumes.

I have ran out of ideas, I'm not sure where else to look.

Any help will be greatly appreciated.

Thanks and Regards
KBR

 

[djr111]

you might want to explain more about what performance issues you are having if you expect to possibly get help.

 

[artfulbodger]

We definately need more info, but.....

I was installing Solaris 10 on a PowerEdge 2850 with a PERC RAID card....it was horribly slow even though there was no disk or cpu activity. I reconfigured it not to use RAID and suddenly everything was fine. I then traced it to a compatability issue with my PERC card.

Try running it in SCSI and not RAID...see if that fixes it

 

[KenCunningham]

You might also want to investigate your mirroring, particularly with regard to possible controller issues.

The internet - allowing those who don't know what they're talking about to have their say.

 

[blarneyme]

A GigE network adapter that is saturated can consume a cpu.

System is not swapping much. Well with 128GB of RAM, and you have 40% free memory, I would ask why you are paging.

Disk I/O isn't bad. What do you consider bad? And you state it peaks on /tmp. /tmp is a filesystem, and if you only have two disks, then most other filesystems will have the same disk as /tmp.

Running a database? Java? Not even a vmstat or prstat. Did this just begin occuring?

I have 7 years on large Solaris servers and 16 years on large AIX servers, so a bit of advice: if you want some direction on where to look, provide some data.