On 11/12/2009 03:42 AM, xinfang wrote:
Thanks, I have used the qpid-stat tools and looked at each queue's speed seems the same, and no one is special. And i can see when each queue byteDepth size exceeds 51m, then broker crushes. so i tired to setup the default queue size to 50MB(qpid default queue limit size is 100M), then it works. i'm just confused that my server's Mem is 8gb ,for qpid 100q * 50M= 5g, my senders need around 10 minutes in order to make queue byteDepth size 50M, so i don't know who uses another 3gb mem. for normal situation when queue byteDepth size get to proximate 70M(100q*70=7g) then broker crushes.
The headers and other overhead are not included in the size reported (and measured for queue limits) at present. That can account for some of the extra memory used. Each connection will also have some buffers associated with it.
another situation i try to run 200 topic under one connection(sender and receiver). if i setup the queue size is 20M then broker also crushes. so i trun to setup size to 10M ,after it runs around four hours,its sender abrots, and i go to use qpid-stat tools checked the queue info. every queue Depte size is not exceeding 10M , but some queue's byteEnqueue size is large (such as 2g) and some queue's byteEnqueue size is small 50M.
Your subscriptions are all identical? I.e. they should all match the exact same set of messages? Are you using any flow control on the subscribers?
A connection is treated as a single stream as far as the IO is concerned and its certainly possible that the sender gets to dominate (one outgoing message may result in 200 incoming messages), so the queue growth is probably to be expected.
By 'byteEnqueue' do you mean the bytes-in field from qpid-stat? What size are your messages? Are you using the default management publish interval?
--------------------------------------------------------------------- Apache Qpid - AMQP Messaging Implementation Project: http://qpid.apache.org Use/Interact: mailto:[email protected]
