Re: Performance Bottleneck?

["Joe Wells" <jwells@xxxxxxxxx>]

Aaron,

Thanks for the tip.  Using the TRCTCPAPP command yielded some interesting
results.  In particular, the following - 

00000040:797896 ap_log_rerror()                                       
00000040:797912 log_error_core()                                      
00000040:798024 log_error_core():  "[Wed May 21 13:02:38 2008] [error]
00000040:928264 ap_log_rerror()                                       
00000040:928288 log_error_core()                                      
00000040:928344 log_error_core():  "[Wed May 21 13:02:38 2008] [error]
-- 05/21/2008 13:02:39 ---                                            
00000040:054808 ap_log_rerror()                                       
00000040:054832 log_error_core()                                      
.
.
.--- 05/21/2008 13:02:40 ---                                             
 00000040:028696 ap_log_rerror()                                        
 00000040:028720 log_error_core()                                       
 00000040:029320 log_error_core():  "[Wed May 21 13:02:40 2008] [error] 
.
.
.
--- 05/21/2008 13:02:52 ---                                            
 00000040:014488 ap_log_rerror()                                       
 00000040:014520 log_error_core()                                      
 00000040:014608 log_error_core():  "[Wed May 21 13:02:52 2008] [error]
 00000040:144808 ap_log_rerror()                                       
 00000040:144840 log_error_core()                                      
 00000040:144896 log_error_core():  "[Wed May 21 13:02:52 2008] [error]
 00000040:265592 cgi_read_stdout.                                      
 00000040:265680 xlate_out_filter() - output data from upstream length 
 00000040:265904 cgi_read_stdout.                                      

Note the first time of 13:02:39 (which is when the RPG completed) and the
last time - 13 seconds!

The first two lines of the log in the IFS (which corresponds to the above
spool file snippet) have this - 

[Wed May 21 13:02:34 2008] [error] [client xx.xx.xx.xx] QzhbCgiParse: No
matching field name found., referer: http://....... 
[Wed May 21 13:02:38 2008] [error] [client xx.xx.xx.xx] QzhbCgiParse: No
matching field name found.ght: bold;, referer: http://.........

And then it has a whole bunch of entries.....but they do not seem to contain
error messages - just HTML that is in the page.

Anyone have any thoughts?

Thanks,

Joe
> > > On 5/21/2008 at 11:05 AM, in message
<1211385927.10614.6.camel@aaronthinkpad>, Aaron
Bartell<aaronbartell@xxxxxxxxx> wrote:
> One thing you could do is turn on verbose mode in Apache and see if
> there is anything in there.  I believe it has timestamps which would
> allow you to determine if it was Apache or the network or the client
> rendering the HTML.
>
> Here is how I do it:
> STRTCPSVR SERVER(*HTTP) HTTPSVR(MYRXS '-vv')
>
> Run a few transactions and make sure the lag occurs.  Then end the same
> named HTTP server instance and do a WRKSPLF QTMHHTTP and you should see
> a handful of spool files that were created.  Note you need to end the
> instance for the spool files to be created.
>
> If you don't want to stop the HTTP server instance you can do a
> TRCTCPAPP command and then use PRTTRC (I believe) to view the results.
> This is what IBM usually asks me to do when I am on 2nd and/or 3rd level
> support trying to figure out an issue.
>
> HTH,
> Aaron Bartell
> http://mowyourlawn.com 
>
>
>
> On Wed, 2008-05-21 at 10:38 -0500, Joe Wells wrote:
> > I have an RPG-CGI web application that would appear to be periodically
> > hitting a bottleneck somewhere along the line. I have tracked the time 
> it
> > enters and exits the RPG and that always results in subsecond response. 
> > Unfortunately, the web page itself does not display that quickly.  
> Sometimes
> > it is subsecond, but frequently it is much more (as much as 10-15 
> seconds!).
> >  We have given QHTTPSVR lots of memory, but that had no effect.  
> So.......I
> > am looking for suggestions of where to look next.  I specifically want 
> to
> > rule out the "i" as the culprit before I contact the network folks to 
> have
> > them sniff around.
> >
> > Thanks,
> >
> > joe