Re: What happened to rpgenerationx.com?

Brad,

One of the questions we've been dealing with is the scalability of a Web
site that deploys CGI programs.  Isn't that something that deserves credible
testing?  But it seems that you're throwing out ideas and numbers without
thinking!

I stress tested a Relational-Web program and an Easy400 program that produce
identical HTML streams.  The Easy400 CGI program remains active (runs in a
named activation group), to optimize performance.

In the case of ten (10) concurrent clients, the CGI workload is divided and
balanced between ten (10) active BCI Jobs.  Unfortunately that's how CGI is
designed.  In this case the Relational-Web workload is queued to a single
RPG program.  An operator can start more instances, if needed.  The JMeter
results prove the Relational-Web architecture to be much more efficient, and
scalable.

Over a local area network, with a constant delay of four (4) seconds, and
over a test period of several minutes, the average Relational-Web response
is about 600 milliseconds, while the average CGI response is about 7000
milliseconds.  The reason for the 4 second delay is mostly for the benefit
of the CGI program.  Otherwise the CGI workload surpasses my CPU capacity.

If I understand correctly, you tested the programs with ten (10) nearly
simultaneous hits over a 128 KB router.  Rather than testing the performance
of the programs, it probably buried my CPU and router.

In the case of the router, the maximum capacity is 16K per second.  It's
probably quite a bit slower when balancing streams across 10 concurrent
connections.  And you were requesting about 240 K worth of HTML.  The data
transport alone probably accounts for 15-18 seconds of your response time?

That's why I asked, are you interested in testing the performance of a
software architecture?  Or more interested in testing the performance of my
CPU and router?

Nathan M. Andelin
www.relational-data.com



----- Original Message -----
From: "Brad Stone" <brad@bvstools.com>
To: <web400@midrange.com>
Sent: Wednesday, July 31, 2002 10:28 AM
Subject: Re: [WEB400] What happened to rpgenerationx.com?


> On Wed, 31 Jul 2002 09:34:58 -0600
>  "Nathan M. Andelin" <nandelin@RELATIONAL-DATA.COM> wrote:
> > Brad,
> >
> > On one hand I'd encourage you to run your tests again.
> > During your test, I
> > may have been running a stress test of my own which
> > overloaded my 128 kb
> > router.
> >
> > On the other hand, I really think you need to try it from
> > JMeter or
> > something like it to get comparable results.  It sounds
> > like your test may
> > have overloaded my CPU and my network.
>
> CPU plays a role, so does network.  But the results are
> comperable to me hitting my own sites on the same machine.
>
> >
> > Do you want to test the performance of my CPU and my
> > network (which are
> > quite limited), or the performance of the software
> > running on it?
>
> Well, they all play a role.  128k isn't peanuts.  And the
> data being returned isn't that much.
>
> >
> > If you were running your tests from the Netshare system,
> > your results may
> > have been skewed by an overloaded system on your side.
>
> No, the system never was over 60% during the test.  I keep
> an eye on that.
>
> > Do you want to test
> > your system or mine?  Use a dedicated Pentium and
> > software with thread based
> > design, and you'll be stress testing my applications, not
> > yours.
>
> But I don't believe JMeter tests the way a browser acts.
> That's why I test the way I have it.  I've had very
> consistant results even when doing 100 simultaneous hits.
>
> I've even load tested my own Linux Apache server running on
> a 56k dialup with excelent results... bandwith is rarely the
> problem.
>
> I'm not trying to prove your product doesn't work.  But, if
> you're that worried about your CPU and network causing
> problems, maybe you have a customer with a bigger box that
> would allow me to test on instead.  I was under the
> impression that your product saved CPU and resources as
> compared to the HTTP servers.. so that should have some
> affect.
>
> Maybe I'll set up my PC to pound on your server for a while.
> :)  We'll see.
> _______________________________________________
> This is the Web Enabling the AS400 / iSeries (WEB400) mailing list
> To post a message email: WEB400@midrange.com
> To subscribe, unsubscribe, or change list options,
> visit: http://lists.midrange.com/cgi-bin/listinfo/web400
> or email: WEB400-request@midrange.com
> Before posting, please take a moment to review the archives
> at http://archive.midrange.com/web400.
>