Network outage 'freezes' router.
2 posts in topic
Flat View  Flat View
TOPIC ACTIONS:
 

Posted By:   Chris_Godwin
Posted On:   Thursday, June 6, 2002 04:30 PM

Hi all, I am experiencing some problems with SwiftMQ on AIX. It appears to me the keep alive messages are failing to detect a 'dead' client, which is effectively freezing the router. This is occuring on SwiftMQ versions 2.1.3 and 3.2.0. I have a simple configuration of a single router, one publisher and several subscribers. If one of the clients network goes down (i.e I pull out the network cable), within a short period of time the publisher, all other clients and the router itself is stopped. The router will not respond to attempts to connect (including the administration tools). If the cable is replaced, the system will recover within a minute or so. If I run the swiftmq router on an NT machine, the 'dead' client is detected and the sys   More>>

Hi all, I am experiencing some problems with SwiftMQ on AIX. It appears to me the keep alive messages are failing to detect a 'dead' client, which is effectively freezing the router. This is occuring on SwiftMQ versions 2.1.3 and 3.2.0.


I have a simple configuration of a single router, one publisher and several subscribers. If one of the clients network goes down (i.e I pull out the network cable), within a short period of time the publisher, all other clients and the router itself is stopped. The router will not respond to attempts to connect (including the administration tools). If the cable is replaced, the system will recover within a minute or so. If I run the swiftmq router on an NT machine, the 'dead' client is detected and the system restored in a timely manner.


I have tested this with SwiftMQ 2.1.3, 3.2.0 on AIX 4.3.3 with IBM JDK 1.3.0 (with JIT enabled and disabled), AIX 5.1 with IBM JDK 1.3.0 and 1.3.1. I have also tested this on linux with Suns JDK 1.3.1 in client and server modes. Each time I observe the same result. I have tried turning off flow control and creating a new listener with a short keep alive interval, increasing the number of threads in the pool, but have not observed a difference in behaviour.


Has anyone else observed a similar problem and have any ideas?


Thanks.

   <<Less

Re: Network outage 'freezes' router.

Posted By:   Oliver_Ortner  
Posted On:   Friday, July 26, 2002 02:39 AM

Hi Chris,


it seems that we have the same problem on AIX 4.3.3 with SwiftMQ 2.1.3! We have a C/S application and whenever the number of client extends (approximately) 200 the system seems to freeze.


We have detected two possibilities for this:



Performance problems as you described in another posting


(Performance reduction in 3.2.0? http://www.jguru.com/forums/view.jsp?EID=906291


or something similar as you described in this posting. The
application runs in a old token ring environment and under heavy load it may seem that the client connection is lost.


Any ideas for this urgent problem?!


THX

Re: Network outage 'freezes' router.

Posted By:   Andreas_Mueller  
Posted On:   Friday, June 7, 2002 12:35 AM

Just tested it here. Connected remote with the Explorer and pulled the cable. The router doesn't freeze but I use JDK 1.4/Linux.


Pulling a cable doesn't drop a TCP connection. It is still up. The keepalive messages are short (25 bytes) and it might be that they are still in the TCP buffer of your machine. It might also be that the socket.write just stucks. It doesn't stuck here.


Outbound writes are performed by threads from the jms.connection pool. The default max is 1 so there is 1 thread. If that stucks at write, the router freezes, of course. You can increase the max to a higher value but you will have then some performance degredation due to thread context switching (max=1 is optimal).
About | Sitemap | Contact