Re: CGIDEV2 performance versus Net.Data

[Mel Rothman <mel@xxxxxxxxxxxxxx>]

Over the course of this discussion, it's become clear that measuring this stuff
is neither easy nor unambiguous and performance is can be more affected by how
tools and facilities are used than by which ones are used.

Joel Cochran's approach isn't necessarily what other CGI users do or should do.

Again, well written CGI named activation group programs should consume a
constant amount of storage over their running life.  And, although Joel
apparently had difficulites with it, one can, and should, throttle the number of
threads the server uses to match the system's configuration in order to maintain
performance under heavy load.  See the Webmaster's Guide for details.

Sites running hundreds of CGI programs have something wrong with their CGI
use..  Some of the common errors are:

- using CGIs where static pages should be used.
- using a CGI where the data change only infrequently (static pages should be
rebuilt by batch processes when the data change.  CGIDEV2's WrtHtmlToStmf
subprocedure makes this easy to program.
- not combining multiple related/similar CGI programs into one program.
- not identifying the most heavily used CGIs (the old 80/20 rule) and running
them in named activation groups (and, where applicable, have related ones share
a single activation group).
- if memory is tight, one might consider running infrequently used CGI programs
in *NEW activation groups in order to conserve storage.  Generally speaking, I
recommend that customers make sure they have more than enough memory.



Mel Rothman
CGIDEV2 Author
IBM eServer iSeries Custom Technology Center (iCTC)
Rochester, Minnesota


"Nathan M. Andelin" wrote:
>
> > From: "Mel Rothman" <mel@rothmanweb.com>
>
> > Where are you measuring Relational Web's CPU?  In the message
> > server?  In the application server(s)? Both?  It seems that both is
> > what should be measured.
>
> Yes, both.  Just to confirm, the CPU time for the Relational-Web message
> server is reflected in the HTTP Server's BCI job.  The CPU time for the
> application server is reflected in a separate job.  There's also a 3rd Job I
> call the "switch" used to route and load-balance requests.  I add that time
> too.
>
> For Easy400 and Net.Data I count CPU time for both HTTP Server BCI and BCH
> Jobs.
>
> > What about response time comparisons?
>
> I don't have a good way to measure it.  Over my 128 kb router it's hard to
> tell ;(
>
> Your other comments Mel, have convinced me that named activation and CGI go
> hand and hand, for high use scenarios.  But Joel Cochran apparently shuts
> down restarts the HTTP server daily to reclaim CGI resources.  I think his
> application consists of about a dozen CGI programs.  What about sites that
> deploy hundreds?  Shut down and restart every couple hours?
>
> Nathan M. Andelin
> www.relational-data.com