Network/CPU utilization sanity check???

[Alan Meyer_4]

I have an internal customer whose application seems to bog down and lose communication connections whenever the CPU utilization get's over 75%. I've been telling him that it sounds like an application problem to me since there are plenty of CPU cycles left over.

Just today he called and said that beginning last night at 2300 hours, he was experiencing the problem again. I began to look into what was listed in the syslog beginning then and have discovered that since 2300 hours last night, roughly 30,000 files have been deleted via a ftp connection. This roughly close to 1 file deleted per second.

Am I off base in saying that there in lies the CPU/communication problem?

[Mark Nieuwboer · Accepted solution]

Hi Alan,

I would say yes.
The ftp connections is causing the 75% cpu utility but thats not the problem. The problem is 30.000 delete commands trough ftp. This is causing many threads on the system and the network. But do you have some measureware data, there you can see how many threads and network probing there is.

grtz. Mark

[Steven E. Protter]

Shalom,

You are right its not the ftp per say causing the problem.

Some things to look at:
A directory in ftp with many files in it. the cpu is involved in providing ls output to the ftp client

The application is clearly doing to many files. The cpu is waiting for i/o to update the lists afer deletes, this is a caused by a poorly designed appliction in my opinon. I'm pretty much behind your concept here.

SEP

[Bill Hassell]

You can verify that the network is OK by running lanadmin and looking at the stats. If you haven't cleared the stats recently, you'll need to calculate ratios since a few errors and retries are normal over weeks of networking.

You need to characterize "lose communication connections" a bit more clearly. Are these telnet or ssh sessions or custom ports used by other computers? If so, do the applications require very tight timing (response) requirements? Is the CPU utilization purely CPU or si there a lot of disk activity? And most important, does sar -v 1 show more than 1 digit of po (page out) activity? 2 digits (or more) indicate a severe lack of RAM so that paging (swapping) is taking place. This will drastically reduce performance for processes that are paged out.

[Alan Meyer_4]

Thanks for your help guys, I'm now off to gather more data.