Re: CGIDEV2 performance versus Net.Data

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

I've pretty much come around to accept that the use of named activation
groups are a key to CGI scalability.  Named activation dramatically cuts CPU
time.

The concern that remains (perhaps only in my mind) is the issue of memory.
We have pointed out that the mechanics are already in place to dynamically
add HTTP Servers BCH Jobs to support the user load, up to a configurable
limit.  Let's say that limit is the system default of 40.  Let's also say
that your application consists of 100 CGI programs.

It seems to me that named activation opens the possibility of 4,000
instances of CGI programs to be running concurrently (100 CGI programs times
40 BCH server Jobs).  Running may not be the correct word.  In fact, they
probably aren't running.  But they have memory and files and other resources
allocated and opened.  The number of program activations could grow from 1
to 4,000 depending on how busy the server was, and how the HTTP Server
routed new requests.  My observation is that the HTTP Server's BCI Job
simply routes CGI requests to the next available BCH Job.

There is nothing similar to the Java garbage collector to deallocate the
resources.  So shutting down the HTTP server appears to be the only
practical way to reclaim those resources.

By making these points, I hope I'm not offending anyone.  I just want to
sort all this out, so I can refine my message, and not do a disservice to
myself or anyone else.

Thanks,

Nathan M. Andelin
www.relational-data.com