1834140 Members
2582 Online
110064 Solutions
New Discussion

Re: Performance

 
Ade Falusi_1
Occasional Contributor

Performance

One of our servers is currently using 96% of memory and 0% CPU idle time. Though recent performance show improvement, but the problem remains.
6 REPLIES 6
Victor BERRIDGE
Honored Contributor

Re: Performance

Hi,
Have you looked for orphaned processes?

Just thoughts...

Best regards
Victor
John Bolene
Honored Contributor

Re: Performance

Have you used top or glance to find out what is using the cpu and memory?
It is always a good day when you are launching rockets! http://tripolioklahoma.org, Mostly Missiles http://mostlymissiles.com
Andy Monks
Honored Contributor

Re: Performance

Just because your memory is almost 100% used, doesn't indicate a problem.

Have a look for large processes using 'ps -el'. Under the 'sz' column you'll see the size in pages (4k).

Obviously if you have glance use that (it's nicer).
Jim Welch
Respected Contributor

Re: Performance

check the buffer cache settings in the kernel also...the default setting (50% dynamic buffer cache) can make it seem like your system is short on memory.
Any sufficiently advanced technology is indistinguishable from Magic - Arthur C. Clarke
Greg Hall
Frequent Advisor

Re: Performance

Unix systems generally will use all memory available. Anything not used for processing will be used for mapping filesystems. Use vmstat to see if any page outs are occurring, see the pi and po columns. Essentially you want 0 in the po column or only occasional low numbers. Anything else generally indicates excessive paging and a memory bottleneck. And the best solution is more memory. Like the others said, Glance is nice if you have it, or ps -el will show you some memory usage per process.
CPU 0% idle: could be rogue processes, look for large numbers in the TIME column in top. but don't jump to conclusions. If you haven't booted for a while some of these are meant to be large. Also try a reboot when convenient, does the problem return? Also, do you have high wait i/o percentage, this can be caused by many things, including excessive paging as above. How high are the numbers in the run queue, use sar -q, or vmstat column 'r'? The recommended level is 2-3, alittle higher is sometimes ok, but consistently high numbers result directly in poor response time.
good luck, Greg.
Confucious Confused
Philip Chan_1
Respected Contributor

Re: Performance

The dynamic nature of buffer cache could used up all memory for FS caching purpose. That is not really a problem because when an application ask for memory then the OS will release some from buffer cache for it. To see a better picture of this, run glance (if you got it) then look at the buffer cache size.

Check swap rate too, as long as the rate is zero or a low number then there is no problem.

Rgds,
Philip