Web Services performance issues, first time inquiry etc.
Last Post 05 Feb 2013 04:19 PM by Scott Klement. 5 Replies.
AddThis - Bookmarking and Sharing Button Printer Friendly
  •  
  •  
  •  
  •  
  •  
Sort:
PrevPrev NextNext
You are not authorized to post a reply.
Author Messages
Chris
Basic Member
Basic Member
Posts:117

--
18 Jan 2013 05:23 AM

One area of concern I am having is that after a period of inactivity the first hit to a service running under one of our Apache instances can have some lag time before it responds. This can also occur if there are more requests than there are active instances; it seems that its a bit tardy spinning up new ones.

 

What can I do to minimize the delay? What do I need to keep from paging out? The entire Apache server and it instances or just the programs fulfilling the requests? 

Ringer
Veteran Member
Veteran Member
Posts:1759
Avatar

--
18 Jan 2013 07:21 AM
I what context? RPG CGI? PHP? Java? And for the number of initial instances, you can set that size in httpd.conf.

Chris Ringer
Chris
Basic Member
Basic Member
Posts:117

--
29 Jan 2013 06:26 AM

RPG CGI.

 

I do set a number of instances, however the system is not quick to respond on first inquiry if a decent amount of time has passed since last use. I am assuming it was paged out. Hence, do I just need to keep the RPG/CGI in memory or the entirety of the apache instances

Scott Klement
Editorial Staff Member
Editorial Staff Member
Posts:16382
Avatar

--
02 Feb 2013 06:08 AM

One common mistake that I've found is that shops basically have IBM i set up for interactive use.   In other words, they have a subsystem (often QINTER) where they do almost all of their processing, so they have their memory pools set up to give QINTER pretty much all of the memory.

As the workload changes, and more work is done in the HTTP server (tho, i guess this philsophy applies to other things too, like SQL) they don't consider the fact that the memory is all allocated to interactive.   

This is particularly true for my customers (I work for Profound Logic Software) where they might transition away from 5250 interfaces all together, and then do the same type of work in the more modern web interface.   They complain that it runs slow, but when you look at it, you see that the problem is just that they have all their memory allocated to the (now unused) 5250 subsystem.  Simply re-allocating it to the  HTTP server makes the performance more or less equivalent.   (Tho, web does use a bit more memory, it's not usually too much -- and memory is cheap, at any rate.)

So....  that's my suggestion.  Take a look at your memory pool allocation.  Maybe even consider adding some memory to the system.  That's usually why you get those long startup times.

Chris
Basic Member
Basic Member
Posts:117

--
05 Feb 2013 08:09 AM
Hi Scott, had you on my system to debug an issue once (ended up you discovering old css or such in an ifs)....

Oh, I have memory pools for the web instances out there.. I tend to group functions into pools, however it seems Apache loves going into *BASE - while it has 36g I probably will need to give Apache its own pool or put it where I keep ODBC and similar loads. I fully admit to heavy QINTER pooling, currently its eating about 105g but it is capped to 26% of the system.

Any ideas of how I can determine memory needs for Apache RPG/CGI? Rough guesstimates? This is mostly a RESTful service (xml in/out). I could always pop it into a pull with a 10% cap and see what the system thinks.
Scott Klement
Editorial Staff Member
Editorial Staff Member
Posts:16382
Avatar

--
05 Feb 2013 04:19 PM

Oh that's you, eh?  Genuine Parts?

The needs really shouldn't be that different from 5250.  You just have to figure out how many simultaneous users you need and make it similar.  I'm no expert on performance adjusting -- but my general advice is, make sure you have enough memory.  I mean, memory is cheap -- why not?

You are not authorized to post a reply.

Acceptable Use Policy