help with my computer?

[bigchris]

I've got a rather big unix box, running AIX.

Its got 10 dual 1.4ghz processors, 40gb ram, 10tb disk (yes terabytes).

I run analytics software on there, and over the last few weeks i've been noticing a very definite performance degradation in run times.

Any ideas where to start looking?

I do the usual in terms of basic housekeeping - killing zombie processes, using nmon etc but nothing seems to be highlighting why the system is slowing down.

Over to you guys!

 

[unhinged]

"vmstat 1" is your friend.

Not sure if that'll tell you anything over and above nmon actually but you might try installing collectd and having a browse over the graphs to see if something's changing gradually over time.

This analysis you're doing, the data you're analysing doesn't happen to be growing by any chance?

 

[Womble]

That's one of the reasons I love UKGSer so much.

Someone on a motorbike newsgroup is having problems running AIX, sidesteps Google and asks for help on the group...

...and gets a serious and useful reply

 

[Magwych]

Try checking how much space is left in /tmp, ( if AIX uses it, do no t forget /var/tmp ).
On our Solaris systems I have serious performance degradations simply because of teh sheer number of files in /tmp and /var/tmp.

If you have lot of files for processes that have finished, then try removing those that are redundant to see if that helps.

 

[bigchris]

"vmstat 1" is your friend.

Not sure if that'll tell you anything over and above nmon actually but you might try installing collectd and having a browse over the graphs to see if something's changing gradually over time.

This analysis you're doing, the data you're analysing doesn't happen to be growing by any chance?

Yea, i'd been looking at vmstat and it was showing a fair few page faults and reclaims. Dont know why they would be there or why they're growing either.

The data does grow, but i'm only using approx half of the 10 tb for storage, 1tb for system and remaining 4tb for processing.

In terms of usage for processors, i've yet to go over 4cpu's or 10gb ram usage. I think the peak was 7 processors when I was simulating 60 different models simultaneously.

 

[unhinged]

Yea, i'd been looking at vmstat and it was showing a fair few page faults and reclaims. Dont know why they would be there or why they're growing either.

You could try turning swap off.

The data does grow, but i'm only using approx half of the 10 tb for storage, 1tb for system and remaining 4tb for processing.

In terms of usage for processors, i've yet to go over 4cpu's or 10gb ram usage. I think the peak was 7 processors when I was simulating 60 different models simultaneously.

Most of my experience is with conventional databases on disk but my thinking was, the larger the data, the longer it can be to process and if the analysis requires a sequential scan that can't be parallelised then you'll reach a speed limit without using all your CPUs.