John Lim Wrote: > If the cluster cannot handle the load, then the queues will just get > longer and longer. It's interesting that mlwmohawk asked about this > because msession is a lightweight session handler without rdbms > overhead. Similarly queuing is used instead of a real database for the > same reasons - > lightweight without rdbms overhead so it scales better. > > So if the cluster would fail talking to a real rdbms, it will work and > scale better using queuing, a more light-weight technology (no need for > indexes, joins, etc - just push and pop). > > I have a reference here which is not the exactly the same as the above > one I gave, but you should get the idea: > > http://www-3.ibm.com/e-business/doc/content/casestudy/43886.html Well, this is not what I expected. I'm surprised there is a name for it. I've done a lot of similar things to remove load from SQL databases. The important aspect of what "queuing" seems to be the study of what data needs to be in the central database in real time. The stuff that need not be in real-time can be pushed off to a different system. I'm not sure this really works for e-commerce because if sales are batched then inventory is not up to date. Because of this, it is possible you will sell what you do not have in stock. Typically, I analyze systems for logically separate functions, and break out the various subsytems. The big bonus comes from when you can find reasonably read-only components, like cataloges. So, instead of one heavily loaded database, you can have a few specialized databases.
-- PHP Development Mailing List <http://www.php.net/> To unsubscribe, visit: http://www.php.net/unsub.php
