RE: Database Maintenance Models

["Nathan Andelin" <nandel@xxxxxxxxxxxxxxxxxxx>]

> In fact Sun, IBM and others promote the Front-Controller
> Pattern which is exactly a Servlet (or a Filter) where
> everything is tunneled through.

I'm not sure "promote" is the right word.  "Highly used" would be more
precise.  The consequence of which is a bottleneck, perhaps.

I recall a vendor who published a Java benchmark patterned somewhat after
the TCP benchmark.  When run on a 24-way iSeries server, the application hit
a ceiling at some insignificant number of transactions, while the CPU
remained mostly idle.

How did they get past the artificial application-imposed ceiling?  They
deployed it over 50 instances of Websphere, which throttled the CPU, and
increased the throughput, but not nearly to the extent they hoped for.  In
contrast, the traditional TCP benchmarks yielded something like 100 times
more throughput on the same server.

> All Web Frameworks i am aware of rely on that pattern.

We discovered this when reviewing Struts and several other frameworks.
Developers beware!

> You get many benefits when using this pattern,
> starting from easy configuration and authentication
> hooks, to a quite clean dispatching mechanism.

Well, developers get benefits.  I'm not so sure about folks needing a
scalable solution.  Actually, dispatching is fairly trivial.  It doesn't
generate significant CPU load.  On the other hand, aren't we discussing the
idea of a single Servlet doing primarily screen generation and database I/O
which would both be synchronized, CPU-intensive activities?

Nathan.