| Author |
Message
|
| Shalini |
 Posted: Tue Nov 23, 2004 6:26 am Post subject: (resolved) Multiple transmit Queue in Cluster? |
 |
|
Master
Joined: 30 Apr 2002
Posts: 224
Location: India
|
Hi,
MQ 5.3 CSD5
Currently we have a hub and spoke kind of infrastructure.
i.e. the center two QM with full repo(Solaris MQ5.3 CSD5)
and other 51 QMGR as a partial repo (Win 2000 MQ5.3 CSD5) for our MQ Cluster
Practically speaking we will have one "Cluster transmit Queue" from the center QM to pass messages to all 51 partial repo QMGR.
1) I donot know if the single "Cluster transmit Queue" can handle 60,0000 messages/day to be transfered from center (full repo) to the partial repo QMGR each messages of 100KB- 500Kb (approx).
Currently our Non MQ Cluster handles this by having transmitq:Senderchannel in 1:1 ratio (as we all know).
2) Or is there any way to have diffrent "Cluster Transmit Queue " for diffrent MQ Cluster for diffrent set of partial QMGR.
Say 3 Partial repo QMGR (part of one MQ Cluster) QMGR connecting to center QMGR so I have 17 MQ Cluster  (yes the center QMGR will be the part of 17 QM Cluster).
By This I also resolve my issue where the images of cluster queue of all 51 QMGR will not be seen in all partial qmgr.
We have 100 Local Queue for each Partial repo QMGR so If I put all in under one cluster I can see images of other 50 QMGR Cluster + my local Cluster queue in my MQ Explorerer
which would be very hard for our support team to monitor.
Please suggest/comment.
Still I donot know if the single "cluster Trasmit Queue" can handle for routing such huge messages ? 
 |
|
| Back to top |
|
 |
| csmith28 |
| Posted: Tue Nov 23, 2004 6:34 am Post subject: |
|
|
Grand Master
Joined: 15 Jul 2003
Posts: 1196
Location: Arizona
|
| Quote: |
| 1) I donot know if the single "Cluster transmit Queue" can handle 60,0000 messages/day to be transfered from center (full repo) to the partial repo QMGR each messages of 100KB- 500Kb (approx). |
That kind of volume shouldn't be too much trouble as long as the host server is robust enough to handle the load and you have enough defined logs and space in your /var/mqm/log/QMGR/active (assuming all the messages are persistent) directory. Preferably /var/mqm/log will be it's own filesystem. See the MQ Administration Guide for full details.
_________________
Yes, I am an agent of Satan but my duties are largely ceremonial. |
|
| Back to top |
|
|
| Shalini |
| Posted: Tue Nov 23, 2004 7:01 am Post subject: |
|
|
Master
Joined: 30 Apr 2002
Posts: 224
Location: India
|
Hi csmith28,
Thanks for your reply.
All messages are persitance.
1) Do I need to do any changes on
| Quote: |
| enough defined logs and space in your /var/mqm/log/QMGR/active |
As current non MQ Cluster is handled without any interruptions.
2) say If I have 17 Cluster defined do I need to higher machine configration then the non MQ cluster system
Please comment |
|
| Back to top |
|
|
| jefflowrey |
| Posted: Tue Nov 23, 2004 7:32 am Post subject: |
|
|
Grand Poobah
Joined: 16 Oct 2002
Posts: 19981
|
You should be very judicious with the use of a cluster. Define clusters for logical environments (dev/test/prod) and for large scale divisions in your organization (regional offices) rather than for small scale groupings (like department or business division).
The complexities of overlapped clusters and the technical challenges of managing and maintaing them start to outweigh the administrative benefits as you add more clusters.
It seems, from stories I've heard, that having a single cluster per application can lead to ... issues...
That said, yes, as you add more queue managers to a cluster, you will use up more resources on a full repository.
Also, with clustering, you don't end up with a "hub/spoke" architecture, unless your network forces that to be the case.
When two queue managers in the same cluster need to communicate, they will automatically define channels between themselves, that do not pass through the FR. They only go to the FR if they need to discover each other, not for routine communications.
So if your FR cluster sender isn't running, and the automatically defined CSDRs on that QM aren't running, that's not necessarily a problem.
_________________
I am *not* the model of the modern major general. |
|
| Back to top |
|
|
| Shalini |
| Posted: Wed Nov 24, 2004 2:32 am Post subject: |
|
|
Master
Joined: 30 Apr 2002
Posts: 224
Location: India
|
Hi jefflowrey
| Quote: |
You should be very judicious with the use of a cluster. Define clusters for logical environments (dev/test/prod) and for large scale divisions in your organization (regional offices) rather than for small scale groupings (like department or business division).
|
We are judicious as we have diffrent QMGR for diffrent geographical location: yes we manage 51 complete diffrent Geographical location
| Quote: |
That said, yes, as you add more queue managers to a cluster, you will use up more resources on a full repository.
|
How to calculate the same.
| Quote: |
unless your network forces that to be the case.
|
Yes that why we use hub and spoke arch
| Quote: |
| So if your FR cluster sender isn't running, and the automatically defined CSDRs on that QM aren't running, that's not necessarily a problem. |
No Its a very big problem as each message is a some time million $ for our client and we have Priority calls based on minutes (not on hours)
How to get rid of messages struck in the "Cluster Queue" when my PR QMGR starts with inactive channels.
 |
|
| Back to top |
|
|
| jefflowrey |
| Posted: Wed Nov 24, 2004 4:24 am Post subject: |
|
|
Grand Poobah
Joined: 16 Oct 2002
Posts: 19981
|
Okay, so the only route between any two queue managers in your cluster is through your hub.
Then, yes, it's a problem if they are sitting around in the S.C.T.Q.
It sounds like you would be better served with a more advanced monitoring tool than MQ Explorer. I think you have a clear business case for spending at least some additional amount of money! If spending $10,000 or $100,000 allows you to prevent one $1,000,000 message from getting stuck... then you have broken even! 
I don't know how to calculate the additional resource useage of adding a single queue manager to a cluster on the full repository. I think it's fairly small, though. I'm not sure if IBM even makes this public.
Are you seeing any log entries when your QM starts up, for why the channel doesn't start?
Is it all dynamic channels, or just some?
You might also be able to write a PCF program that would start all dynamically defined channels, and run this as part of startup. But I'd try to figure the problem out first.
Have you opened a ticket with IBM yet?
_________________
I am *not* the model of the modern major general. |
|
| Back to top |
|
|
| PeterPotkay |
| Posted: Wed Nov 24, 2004 5:33 am Post subject: |
|
|
Poobah
Joined: 15 May 2001
Posts: 7722
|
We have overlapping clusters, and on Saturday mornings, we send about 50,000 messages in Batch mode between the 2 clusters. The messages are between 25 and 50K each. The common S.C.T.Q. on the Gateway QM does back up slightly every once in a while, maybe up to about 50-100 messages, but it goes right down.
When we tried to run 2 batch jobs at the same time, the S.C.T.Q. stayed over 100 pretty consitently, until one of the jobs ended. But everything kept flowing. Done in 30 minutes.
Note that there were other messages going thru at the same time.
Windows 2000 machine with 2 CPUs and 4 GIG RAM. Kinda weak machines and they still did pretty good.
_________________
Peter Potkay
Keep Calm and MQ On |
|
| Back to top |
|
|
| Shalini |
| Posted: Wed Nov 24, 2004 9:02 pm Post subject: |
|
|
Master
Joined: 30 Apr 2002
Posts: 224
Location: India
|
Hi PeterPotkay/jefflowrey,
| Quote: |
Are you seeing any log entries when your QM starts up, for why the channel doesn't start?
|
No
| Quote: |
Is it all dynamic channels, or just some?
|
All are Dynamic cahnnels nothing else |
|
| Back to top |
|
|
| jefflowrey |
| Posted: Thu Nov 25, 2004 6:05 am Post subject: |
|
|
Grand Poobah
Joined: 16 Oct 2002
Posts: 19981
|
Are you seeing the problem with every dynamic channel, or just with some of the dynamic channels?
_________________
I am *not* the model of the modern major general. |
|
| Back to top |
|
|
| Shalini |
| Posted: Fri Nov 26, 2004 7:54 am Post subject: |
|
|
Master
Joined: 30 Apr 2002
Posts: 224
Location: India
|
Hi,
We are seeing the problem with some of the receive channel of Full Repo.
So its some of the dynamic channel
|
|
| Back to top |
|
|
| jgalvin03 |
| Posted: Tue Nov 30, 2004 3:11 pm Post subject: |
|
|
Newbie
Joined: 21 Jun 2004
Posts: 3
|
Hi,
I experienced a similar issue with a channel not starting within an MQ Cluster. I altered the qmgr\channels properties on the Windows side to adopt NewMCA. This remedied my problems and my channels start immediately.
JG |
|
| Back to top |
|
|
| Shalini |
| Posted: Tue Nov 30, 2004 10:30 pm Post subject: |
|
|
Master
Joined: 30 Apr 2002
Posts: 224
Location: India
|
Hi jgalvin03,
Thanks for Reply.
AS per I understand the "Adopt New MCA" is to adopt the new connection of sender channel when the n/w between the sen and rec channnel goes down.
My problem is when QMGR (PR) comes up the receiver of the QMGR should also come up.
But any way as you suggested I will enable it and test into my local system and let you know if the issue is resolved
|
|
| Back to top |
|
|
| Nigelg |
| Posted: Wed Dec 01, 2004 12:27 am Post subject: |
|
|
Grand Master
Joined: 02 Aug 2004
Posts: 1046
|
| Quote: |
My problem is when QMGR (PR) comes up the receiver of the QMGR should also come up.
|
Why?
A channel willstart at qmgr restart if there are any msgs in the xmitq, otherwise it will not. |
|
| Back to top |
|
|
|
|
