> > Currently my pings to the Library of Congress from the UK is 78ms.
> > However, we aren't talking about single (32 bytes?) ICMP
> packets for
> > SRW/SRU, but rather a chunk of XML over http which would typically
> > take several TCP packets to send (so we would be looking at some
> > multiple of 36ms/78ms plus overhead caused by different
> packets taking
> > different routes and packet re-assembly). So I would be surpised if
> > typical network latency of an SRU request to the Library of
> Congress
> > from the UK was less than a few hundred ms.
> From DK, I ping LOC in approx 105 ms . If I choose an
> infrequent term, SRU search takes 250-500 ms (includes SRU
> handling + search at the target). Had it been based on Axis
> it would have been 700-950 ms.
Right, so Axis is now adding a 100%-200% overhead which whilst not
particularly good, isn't as bad as the 1000% hit indicated by Ralph's
measurements. With a (more realistic) more frequent term, or more
complex query I'd hazard the overhead would be around 50-100%. Again not
earthshatteringly efficient but no where near as bad as the earlier
figures suggested.
Would be interesting to see how .Net et al perform against this too...
Matthew
|