I am aware of production systems in use today that use IBM
WebSphere and their MQ implementation of JMS to transfer up
to gigabytes per transaction, tens of millions per day.
Use of multiple servers in that scenario, along with a
whole shi--, er, cartload of additional infrastructure,
is /de rigueur/ in such a scenario.
This is just the sort of thing I wanted to hear. Being relatively new
to java I have not heard of such things but I wondered if this was
just because I am new to the game.
the physical infrastructure and the balanced deployment of
multiple disparate components will dominate your considerations.
Yes, I think so too. I am trying to get the people where I am to see
this also. Unfortunately there seems to be little interest in the
physical structure at the moment. But the one thing no-one can deny is
that we are talking about very large volumes of data.
We speak not yet of redundant systems, uptime, failover,
clustering and other technology matters, much less of the
human-resource considerations of obtaining the services of
competent operations personnel and the concomitant costs.
I worked on one other project where JMS was used but IMHO JMS was
inappropriate on that occasion. The volumes of data were only modest
and the servers were using request-response, i.e synchronous
communication. Once we were using JMS, infrastructure support (or the
lack of it) became a real issue. Queues do seem to require quite a bit
of admin, from where I'm standing.
The admin is a big down side AFAICS. This is why I am not sure it is
worth it if only one queue is used. A single queue means there is very
limited opportunity for parallelism, which could make the queue a
bottleneck. This is why I have multiple queues in mind. I am not sure
how many JMS servers our system will have because the JMS
implementation has not been chosen yet. But it will probably be
activeMQ. I presume that multiple queues will be dealt with using
multiple servers. I am keen to get load balancing since we have a
powerful solaris machine with several CPUs.
I am thinking about dividing the day up into zones and using groups of
queues per zone. This is a bit like dividing a filesystem up into
partitions so each can be sized and managed separately rather than
having one humungous filespace. Does this sound reasonable?
Regards,
Andrew Marlow
