BTW if you put the ActiveMQ broker inside the same JVM as the Lingo server, I'd expect furher improvements in performance; as we can avoid the marshalling on/off the wire etc.
On 10/19/06, James Strachan <[EMAIL PROTECTED]> wrote:
If the only change you made between the 200 and 4 times was move the client to another box, then that drop is the latency overhead (and context switching) cost of using real TCP rather than using the loopback adapter. You might find turning nagler on/off could help reduce latency with TCP (i.e. do you wait for packets to be full or send 'em straight away etc) http://incubator.apache.org/activemq/configuring-wire-formats.html (the tcpNoDelayEnabled option) On 10/19/06, Bernhard2 <[EMAIL PROTECTED]> wrote: > > Hello James, > > > Are you running the broker in the same JVM as the server? That helps > > reduce the latency > > First I had this scenario: > > Machine A: Lingo Server, Lingo Client and ActiveMQ in three JVMS > => throughput 200 Lingo request per second > > Then I had this scenario: > > Machine A: Lingo Server and ActiveMQ in two JVMS > Machine B: Lingo Client > => throughput 4.5 Lingo request per second > > So running an own broker in the server JVM to reduce the latency is a good > idea. But in both scenarios I had the Lingo server and ActiveMQ on the same > machine and the decrease from 200 to 4 requests can not be a problem between > server and ActiveMQ. Only the client moved to machine B. Do you agree? > > Regards, > > Bernhard > > -- > View this message in context: http://www.nabble.com/ActiveMQ-Lingo-performance-tf2473875.html#a6899515 > Sent from the ActiveMQ - User mailing list archive at Nabble.com. > > -- James ------- http://radio.weblogs.com/0112098/
-- James ------- http://radio.weblogs.com/0112098/
