[ic] HELP performance problem on live system

[Bill Carr]

On Thu, 2002-03-07 at 19:33, Peter Jakl wrote:
> >> >I have interchange running in "high traffic" mode that is choking.
> It is
> >> >using mysql.
> >> 
> >> Administer CPR immediately.  Or did you have another definition of 
> >> "choking"?  :-)
> >> 
> >> >I'm running on a 512mb system with a single processor Pentium iii
> >> >800mhz. The system is taking about 25 orders per hour and their
> appears
> >> >to be about 20 to 30 simultaneous connections at any one time. load
> >> >average on the Linux system is around 8.00.
> >> >
> >> >Should I up the maxservers on interchange (currently 10)?
> >> >
> >> >Any suggestions would be greatly appreciated.
> >> >
> >> >Peter
> >> 
> >> For hardware, 512mb is like $140 these days and a U160 hdd isn't much
> 
> >> more.  For software, you can try more maxservers if you're not maxing
> out 
> >> your RAM already (don't count too much for hdd caching).  Also try
> rpc 
> >> mode.  You can index your database really well, too, and don't forget
> to 
> >> regularly expire your sessions.  You can optimize your apache (lots
> of >>docs 
> >> on that).  There is also a lot to do for optimizing the OS (new
> kernel, 
> >> disable "last accessed" on file systems, etc.), but focusing more on
> IC >>is 
> >> probably a better use of your time.  Good luck,
> 
> >That sounds to me like plenty of hardware for 25 orders an hour.
> >Something must be very "unoptimized".  Or you have an overly complex
> >catalog (work on caching).  20-30 connections is not a lot.
> >
> >Start by looking at the logs.  Database, web server, ic.
> >
> >And top, swap, ps ax.
> >
> >What does load avg go down to when you kill IC?  Is this
> >really IC or something else?   Procmail and spamkiller
> >repeatedly bouncing a 120Mb log file comes to mind....
> >A load avg that high might also mean the effects are 
> >snowballing.
> >
> >Memory is cheap, yes, that would help.  You are not using some
> >old slow IDE hard drive are you?  That could ball you up bad.
> >man hdparm.
> >
> >If your machine is that busy, it's unlikely you can afford
> >to toubleshoot it hot.  Bring up another machine first.
> >
> 
> Thanks for the input. I have a SCSI subsystem. Killing IC brings it down
> to virtually idle. There is nothing else impacting performance that I
> can see. I am using ps, top, etc.
> 
> I don't like to throw hardware unless I see that it is definitely
> thrashing. It seems it should not be crawling this badly.
> 
> Each instance of interchange is using 14mb+ of memory which seems
> excessive but then again it is perl and I expect it to be a memory hog.
I am also having trouble with performance. I have a pIII 450, 256RAM,
RAID5, Debian woody, Perl 5.6.1, Mysql 3.23.47 (on seperate host),
Apache 1.3.22, IC 4.8.3

I test with:
	 ab -n 100 -c 5 -C 'MV_SESSION_ID=2eTSjKxz:192.168.1.7' <URL>

I created an index page with no IC tags on it. When I run ab with a URL
to pull the page from IC I get about 2 requests/sec. When I run ab with
a URL to pull the same index page from Apache directly I get about 500
requests/sec.


-- 
Bill Carr
Worldwide Impact
bill@worldwideimpact.com
413-253-6700