| |
|
RSS Feed - WebSphere MQ Support
|
RSS Feed - Message Broker Support
|
  |
|
| HA & SPOF design |
« View previous topic :: View next topic » |
| Author |
Message
|
| boydl |
 Posted: Mon Apr 11, 2005 8:53 pm Post subject: HA & SPOF design |
 |
|
Novice
Joined: 05 Feb 2003
Posts: 16
|
Hi guys,
I know that this topic have been discussed many times b4, however, I still am going to post it  . Appreciate your patiences.
We r planning to implement a set of qmgrs which is fronted w a gateway qmgr. This gives us load balancing for the app qmgrs. Furthermore, with MQ Clustering exits, we shld hv more control to which qmgr that the message is routed to (rather that use the default round robin). However, with this config, there is a SPOF in the gateway qmgr. We understand that we can use OS clustering for Failover, but there will still be the few minutes of downtime for the fail over activities to complete. Furthermore, I doubt that it will be a fully automated process. To get around this, there are suggestion to use 2 gateway qmgr which is fronted by BigIP. What this will do is that it will provide a virtual IP to the application but uses the actual IP to communicate w the gateway qmgrs. Does this design makes sense,i.e. application specify the qmgr, hostname etc. but the actual location or IP is provided from the BigIp? Also, can a cluster contain 2 qmgr with the same name?
To get a reply back, will setting the Reply_toQ and Reply_toQmgr solve the problem?...What i mean is that, in a load balance environment, the qmgr does a round robin on which queue to put the data on. Is it possible to control where the data will be going
Appreciate that you will bare with me... since in my role, I dont have access to the system, so I cant "test" this approach and need to taps on this user group!
regards |
|
| Back to top |
|
 |
| sebastianhirt |
| Posted: Mon Apr 11, 2005 9:52 pm Post subject: |
|
|
Yatiri
Joined: 07 Jun 2004
Posts: 620
Location: Germany
|
| Quote: |
Also, can a cluster contain 2 qmgr with the same name?
|
Hi,
If you try having 2 QMs with the same name in the same cluster, the cluster will be confused and is likely to serve neither of them.
I would advice, not trying to have 2 QMs with the same name in the same cluster.
| Quote: |
| We understand that we can use OS clustering for Failover, but there will still be the few minutes of downtime for the fail over activities to complete. |
You won't have downtime when using an Active/Active solution rather then Active/Passiv.
| Quote: |
| Furthermore, I doubt that it will be a fully automated process. |
I think Active/Active is fully automated.
| Quote: |
| What i mean is that, in a load balance environment, the qmgr does a round robin on which queue to put the data on. Is it possible to control where the data will be going |
Not as far as I know. To have that functionallity, I think you have to look into Work load exits.
My 2 Cents
Sebastian |
|
| Back to top |
|
|
| PeterPotkay |
| Posted: Wed Apr 13, 2005 1:37 pm Post subject: |
|
|
Poobah
Joined: 15 May 2001
Posts: 7716
|
If you have 1 Gateway Qm (GWQM) into your cluster, then it is a single point of failure. Period. Active/Active versus Active /Passive has nothing to do with it.
GWQM running on node1 while node2 does nothing but wait for a failure of node1 is Active/Passive.
GWQM running on node1 while QMX runs on node2 is Active/Active.
Either way, if node1 or GWQM has a problem, you have an outage until (and IF!) GWQM comes up on node2.
I feel your pain. We have a few QMs in our cluster, housing important things like our WB-IMB brokers, and all incoming traffic goes through that one gateway QM. If it goes down, we effectivly lose all those QMs in the cluster too, since nothing can get to them.
How to makle it better? Well for one, do NOT use Active/Active. You want that hardware cluster supporting GWQM as stable as possible. You don't need some problem on QMX on node2 corrupting node1.
Having multiple GWQMs only spreads the problem out. If one of you GWQMs goes down, only the SNDR channels to it are effected. But it is still an outage for those channels.
One solution is to get rid of the need for the GWQM. You won't have a single point of failure if you don't have a single point of failure!!! The only way to do this is to put all your QMs in one giant cluster. No need for a GWQM.
The BigIP is not workable. QMA (not clustered) wants to talk to GWQM1. If you have an IP load balancer sending messages from QMA to GWQM1 and GWQM2, you will have nothing but sequence number errors on the channels. QMA sends message #1 to GWQM1, which is expecting #1. Message #2 goes to GWQM2 (cuz of the load balancer), but wait a minute! GWQM2 is expecting message #1......
_________________
Peter Potkay
Keep Calm and MQ On |
|
| Back to top |
|
|
|
|
|
| |
|
Page 1 of 1 |
|
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum
|
|
|
|
