Performance Tuning

[Grayh]

Hi All,

Where can I find the performance Tuning commands...

The first thing I have checked was iostat but could not get anything from it

# iostat

device bps sps msps

c0t0d0 0 0.0 1.0

then

#netstat ----> I have noticed some TIME_WAIT

Do I have any specific options to use with these commands

What are the other Perf. Tuning commands and what do I have to check in them...

[Ivan Krastev]

You can check disk/process/memory usages with glance,sar, iostat, vmstat.

regards,
ivan

[Adam W.]

grayh,
What exactly are you looking for? There are tons of things to use, glance, top, iostat, netstat, sam (not reccommended, swapinfo......the list goes on and on. What exactly are you looking to check for?

[Adam W.]

sar is another good one. But again it depends on what your looking to do. What version of HP-UX? veritas or onlineJFS?

[Grayh]

HP UX 1i V3... I do not have online jfs..

In a general senerio if a command is taking too long to run or if the performance of the server is very slow..

what are the things I should look from the OS side and what are the things I should look in from the network side...

[OldSchool]

"In a general senerio if a command is taking too long to run or if the performance of the server is very slow.."

the problem is there aren't "general" solutions. I would look at different things for a problem with a "long-running database query" then for "slow to login" issues for example.

[Tim Nelson · Accepted solution]

As mentioned there are performance monitoring tools, sar, top, vmstat, but glance is your best bet (must purchase).
Statistics MUST be reviewd over time, a single point in time knee jerk reaction may cause changes to be made that are not in the best interest overall.

There are no performance tuning commands, there are system configuration commands, each relating to an area of concern, CPU, MEM, or DISK.

Some tuning, but not all, is done with kernel parameters. kmtune/kctune to list them all. There are very few commands that say "go faster". Misconfigured parameters, storage or too little RAM are typically to blame. Patches too are important, keep your system up to date.

Most applications provide guidelines of recommended system configurations/patches.

A proper system setup with approprite sizing for the applications you intend on running as well as an intelligent scheduling of workload is the key to success.

-Directly in response to the iostat output. At the milisecond this was run, nothing was happinging on your one disk.

-Directly in response to TIME_WAIT in netstat. This is a connection that was but no longer exists. Has absolutely no effect on performance.

[Grayh]

Thanks everyone

[rick jones]

I think it would be more accurate to state that any one TCP TIME_WAIT endpoint's effect on performance is epsilon, and that it takes thousands and thousands of them to be meaningful.

WRT other networking things, that too is a very broad area. Personally I start by looking for retransmissions over intervals - snapshot netstat -s -p tcp over an interval and subtract one from the other with beforeafter:

ftp://ftp.cup.hp.com/dist/networking/tools/

Of course that does raise the question "what is an acceptable retransmission rate" and that is another "it depends things"

I also check for lost frames and errors in lanadmin statistics and go from there to consider latency to remote destinations and window sizes and such...