Hi, I'm designing a system that will have a few central services and upward of 6000 processes on thousands of machines on a LAN. They all communicate using ActiveMQ, with the central services having well-known queue names and the distributed components having randomly-generated queue names. I recently profiled the broker and found that every queue results in four threads (one for the queue, one for checkpoint (?), and two for each of two related topics (consumer and producer)). Additionally, every connection to the broker results in two threads (dispatcher and transport). That means that each of the 6000 distributed components will result in six threads, or 36,000 threads total in the broker. I don't know whether using a network of brokers will allow me to reduce this to a reasonable number per process. I also have to worry about the 6000 TCP connections to the broker, which could also presumably be distributed across a network of brokers.
More generally, though, I'm concerned that ActiveMQ, and really JMS, was never intended to support this many clients. Perhaps I should use JMS only for communication amongst the central services, and all communication to/from the distributed services should use a simpler RPC scheme. Does anyone know if ActiveMQ can handle this many clients? What's the most number of clients anyone has seen on ActiveMQ or JMS? Thanks, Lawrence -- View this message in context: http://www.nabble.com/6000-ActiveMQ-clients-tp19983973p19983973.html Sent from the ActiveMQ - User mailing list archive at Nabble.com.
