Hi All, On the Java side, 5M would not be an issue - assuming you had a heap for your broker which was at least say 12M (we'd recommend 3GB ideally as the broker is effectively heap limited). It will be read into memory and stored there until completed delivery. We don't have any ability/config for chunking afaik.
Transient vs persistent messaging would be a key differentiating factor in terms of how well the broker copes (jn terms of both speed and general broker performance under load) - the write to persistent store (if you use one) will slow down delivery end-to-end. Hth - let me know if you need any more info, Regards, Marnie On Mon, Oct 27, 2008 at 10:11 AM, Gordon Sim <[EMAIL PROTECTED]> wrote: > Danushka Menikkumbura wrote: > >> >> Danushka Menikkumbura wrote: >>> >>>> Does Qpid have a mechanism for sending large date sets in chunks?. >>>> >>> >>> How large is 'large'? >>> >>> Say 5M?. Anyway it is up to the sender to decide on this. Yeah? >> > > You can send messages pretty much as large as you like; how well the broker > copes is of course a good question (and depends largely on the size of the > message and the memory available). > > The c++ broker can handle receiving of such messages without having to > store the whole thing in memory. Unfortunately on the sending side it can no > longer do this and the whole message will be read into memory before its > sent off to a subscriber. For 5M that probably isn't a huge issue though[1]. > > Maybe someone else can give the latest state on the java broker for this. > > AMQP 0-10 also supports sending a large logical messages as a series of > separate transfers. We don't currently implement that feature though. I > created a Jira for that: https://issues.apache.org/jira/browse/QPID-1402 > > [1] I did a quick test of a 6M message which worked fine. If you don't have > a store loaded you need to disable the 'staging' mechanism which has a 5M > limit by default (--staging-threshold 0). I think I'll change that to be the > default as the sending side of this no longer has the desired effect anyway. >
