On 07/13/2011 07:21 AM, John Toohey wrote: > Maybe "enterprise" is the wrong word. I have a system with multiple > servers, some in Java, some in Clojure, and the Console servers > developed in Seaside running on Pharo. I need to have all these > servers talk to each other, and today I'm using JMS over ActiveMQ to > do that. I need to use Stomp as a protocol to allow my Console > servers to push data updates to the main servers, and its fragile. It > could be that my implementation is wrong, and you could advise on > that. What is your experience in pushing messages from Smalltalk VMs > to Java and C servers? Any help would be most appreciated. (My systems > supports 200 end to end transactions per seconds, with many thousand > concurrent transactions. It's deployed on 16 servers in an EC2 cloud)
With that level of concurrency you may want to avoid solutions based on FFI. FFI calls block the entire VM, there was a recent SqueakDBX thread discussing this. Plugin based solutions don't have this limitation but you run into issues of mapping native threads to Pharo processes. That in general makes socket based solutions preferable because they don't share these limitations. Since HTTP is my hammer and assuming you don't need message persistence and use messaging more or less for synchronous communication (maybe I misunderstood you) have you considered using HTTP and a load balancer? Cheers Philippe
