> > Average response time for the queries is 445ms.
> > [...]
> > I then tried making the SRW message by hand, similar to building the
> > SRU URL.  [...] Average response time is 46ms.
> 
> Woah!  Dude!
> 
> You're saying that the toolkit you were using to build your SRW XML
> packets was taking ten times as long as everything else you're doing
> _including network overhead_?  Man, that is one deeply 
> doo-dooey toolkit.

This is probably Axis. Axis is not a particularly efficient engine, not
least of all since they got carried away with the class injection
pattern...

However this is really an analysis of the efficiency of Axis rather than
SRW or SOAP per se (and I don't know if Ralph attempted any tuning of
Axis)/

Although 445ms is quite a large latency, 46/36ms indicates a pretty (and
atypical) fast network and database. What Ralph appears to doing here is
high performance clustering over a low latency network, and SOAP has
never claimed to be a good interprocess communications protocol for HPC
over low latency connections (whatever the GRID community may try to
lead you to believe).

In a more typical application, the network latency and database
responses would I believe be an order of magnitude higher than the
latency introduced by the SOAP engine (even one as slow as Axis).

Matthew