At 19:51 20-04-2004 +0100, Robert Sanderson wrote:
>One other point I'm going to make is that in terms of computational
>resource usage, SRW servers will be much lighter weight than Z39.50. Many
>of the overload issues that content providers have now with Z will be
>alleviated by exposing SRW instead. Why? Because, no offence to any
>Z implementors and this applies to our server just the same, Z39.50
>servers are not well written in terms of high load situations. Apache/IIS
>on the other hand -have- to stand up to hundreds or thousands of
>concurrent users every day all over the web.
There are elements of truth to this, but I sure wouldn't hold my breath for
any ground-breaking performance boosts resulting from a migration from
Z39.50 to SRW. There are lots and lots of ways to write inefficient or
poorly scalable database code that have nothing whatsoever to do with the
actual on-the-wire protocol. If a local library database engine is scaled
to handle the workload of a library staff of five and a few dozen local
internet users, then it will easily crumble and die if it's suddenly hit by
metasearch queries from a statewide search portal. That's a scalability
issue that has nothing to do with the carrier protocol.
I'd say that in my own experience, the majority of performance issues we
run into with other people's software have more to do with poorly scalable
database code than with network software.
Sebastian Hammer, Index Data <http://www.indexdata.dk/>
Ph: +45 3341 0100, Fax: +45 3341 0101