Re: [WEB400] What happened to rpgenerationx.com?

["Nathan M. Andelin" <nandelin@xxxxxxxxxxxxxxxxxxx>]

> From: "Andrew Borts"
> We actually worked with the IBM folks to try and resolve it,
> and set thread count accordingly to their suggestions - which
> varied greatly.

In my experience, the thread requirement is roughly 2-3 times the number of
concurrent users.  300 users would need 600-900 available threads, for
example.

Activating that many threads however poses a big problem for the CGI
interface.  Eventually 900 instances of every CGI program become active, and
remain active, if (like Net.Data) they run under a named activation group.

It's like a perpetual memory leak.  Eventually, system resources become
exhausted, and the only solution is to reboot the HTTP Server.

> but the HTTP server isn't as reliable, and darn close
> to sub-par in it's reliability.

Folks need to distinquish between the HTTP Server and the CGI Interface.
While the HTTP Server is both scalable and reliable, CGI has its limits.  It
takes little system resources to support 900 HTTP Server threads.  But 900
BCI (CGI support) Jobs is a whole different matter.

> Is fiddling with the controls of the dang thing the
> option?  Is switching to Apache the option?

The Apache based server uses an equivalent CGI interface, so Apache doesn't
solve the problem.  Just because something works great with five (5) users,
doesn't mean it will work with five hundred.

IBM's answer is a Servlet engine that never instantiates more than one
instance of a Servlet.  The engine just synchronizes requests to one Servlet
rather than creating 900 separate instances of the same.

The same thing can be done with ILE, of course.  But developers need to be
aware of the problem in order to address it.  I addressed the problem with
the design of Relational-Web, which is an ILE framework.

Nathan M. Andelin
www.relational-data.com