Hello MQ community users,

I have a bank customer who is curently using an old (obsolete) Wintel based MQ infra (MQ ver 5.3 to MQ v8.x) and wants to deploy a new MSCS cluster where all stand alone based queue managers from different domains will be hosted (after being migrated).

This old Wintel MQ infra serves 24 x 7 banking applications and connects with queue managers on other platforms (AIX, z/Os).
The reason for maintaining the old MQ infra up to now, was the lack of MQ skills & expertise.

Customer's main scope for the deployment of MSCS MQ HA cluster is to minimize MQ apps downtime and to have a safe patching window (as part of security audit process), which is not currently feasible.

The solution options considered are either MSCS cluster Active/Passive (A-P) or Active/Active (A-A).

I would like to ask the following:
1. Which is the preferable MSCS cluster option (A-P or A-A) to be deployed in terms of MQ Service, objects & meessage availablity?

2. Is it recommended to deploy both MSCS cluster nodes as VMs (virtualized) instead of physical?

3. Is it feasible to combine MQCluster (shared cluster queues on multiple queue managers) with MSCS Active/Passive or Active/Active? If yes, which are the pros and cons?

4. Is it recommended to deploy MSCS as "geocluster" splitting cluster nodes in separated sites (PROD & DR) with their owned shared storage given that bank has already a dark fiber interconnection between both sites (4 lines x 128 Gbps) using storage reploication (VPLEX)?

P.S.
I have reviewed relevant IBM Knowledge Center articles, but i need more customized responses for this particular case:

https://www.ibm.com/support/knowledgecenter/SSFKSJ_9.1.0/com.ibm.mq.pla.doc/q017161_.htm -> Design Clusters

https://www.ibm.com/support/knowledgecenter/SSFKSJ_9.1.0/com.ibm.mq.con.doc/q017820_.htm -> HA configurations

Any advise will be much appreciated.
Cheers Nick.

------------------------------
NICK DAKORONIAS
------------------------------

1. Which is the preferable MSCS cluster option (A-P or A-A) to be deployed in terms of MQ Service, objects & meessage availablity?

Active-Active has the benefit of giving you constant proof your software is functional on both nodes. Not fun to find out when you most need it that your passive node can't do the job. 
Capacity planning is important when using Active-Active, the node better be able to handle all the work at once.
I don't see any difference between Message Availability between A-P or A-A. If a message is on a queue and the queue manager needs to failover, that message is N/A until the QM comes back up and it makes no diff A-P vs A-A.
IF the apps are written to concurrently take advantage of multiple Active-Active queue managers, then the ability to send NEW messages is much higher in an A-A solution (or multiple A-P solutions).



2. Is it recommended to deploy both MSCS cluster nodes as VMs (virtualized) instead of physical?
Depends on your site requirements for physical versus virtual. Be sure you can deploy MSCS in general on virtual in your environment. And make sure it makes sense. What's the point of MSCS? Is it still worth if if both nodes are virtual servers on the same storage, on the same ESX cluster, on the same everything? What exactly are you looking to mitigate with MSCS? If you can still achieve that using MSCS on virtual?


3. Is it feasible to combine MQCluster (shared cluster queues on multiple queue managers) with MSCS Active/Passive or Active/Active? If yes, which are the pros and cons?
Yes.
2 separate concepts. Evaluate them on their own merits.
Used together you can increase availability of the overall solution. Search for IBM presentations on MQ High Availability, there are some with great diagrams comparing the typical H.A. solutions and what they solve and don't solve when used together or independently.


4. Is it recommended to deploy MSCS as "geocluster" splitting cluster nodes in separated sites (PROD & DR) with their owned shared storage given that bank has already a dark fiber interconnection between both sites (4 lines x 128 Gbps) using storage reploication (VPLEX)?
A question that is not MQ specific. The MSCS experts in your company should easily answer this. If they can't, back away slowly and consider a different solution.


As I read your problem description, this seems like a great use case for the MQ Appliances, especially if you are going to use physical Windows servers (phew, I though something smelled funny in here)

Thanks Peter for your prompt response.

Hereunder are my comments and further queries per bullet point  for your consideration.

1.  I Agree with your perspective, but if i am not mistaken, in A-A scenario you have more (quick) downtimes for failover of queue managers, compared to A-P, since queue managers are "splitted" on both cluster nodes and this is a factor to be considered as well. Correct?

2. I agree but in case of virtualized geocluster (see #4), each cluster MQ node will reside on separated ESX cluster and storage, so there will be no single point of failure  on Hypervisor/Storage, but just a failover downtime . Furthermore, there is huge gap in cost between physical & virtual cluster nodes.
The goal is to achieve quick failover of "MQ Service" and ensure a reasonable maintenance window for patching in a cost effcient way.
What do you think?

3.  Do you think that MQ cluster usage will affect existing applications with message affinity (target specific host/QM/Q) ? 
 Is there a need for all apps to be recompiled in order to be cluster aware (remember MQ infra  is currently stand-alone based)?

4.  According to my feedback, Servers/Storage team of the organization, has implemented already "geocluster" with physical servers, for the purposes of an SQL always-on cluster solution. So, they have the expertise to do so.

Rgds, Nick.
     ​

------------------------------
NICK DAKORONIAS
------------------------------

Original Message:
Sent: Fri May 29, 2020 08:27 AM
From: Peter Potkay
Subject: MSCS Cluster deployment options for MQ High Availabilty

1. Which is the preferable MSCS cluster option (A-P or A-A) to be deployed in terms of MQ Service, objects & meessage availablity?

Active-Active has the benefit of giving you constant proof your software is functional on both nodes. Not fun to find out when you most need it that your passive node can't do the job. 
Capacity planning is important when using Active-Active, the node better be able to handle all the work at once.
I don't see any difference between Message Availability between A-P or A-A. If a message is on a queue and the queue manager needs to failover, that message is N/A until the QM comes back up and it makes no diff A-P vs A-A.
IF the apps are written to concurrently take advantage of multiple Active-Active queue managers, then the ability to send NEW messages is much higher in an A-A solution (or multiple A-P solutions).



2. Is it recommended to deploy both MSCS cluster nodes as VMs (virtualized) instead of physical?
Depends on your site requirements for physical versus virtual. Be sure you can deploy MSCS in general on virtual in your environment. And make sure it makes sense. What's the point of MSCS? Is it still worth if if both nodes are virtual servers on the same storage, on the same ESX cluster, on the same everything? What exactly are you looking to mitigate with MSCS? If you can still achieve that using MSCS on virtual?


3. Is it feasible to combine MQCluster (shared cluster queues on multiple queue managers) with MSCS Active/Passive or Active/Active? If yes, which are the pros and cons?
Yes.
2 separate concepts. Evaluate them on their own merits.
Used together you can increase availability of the overall solution. Search for IBM presentations on MQ High Availability, there are some with great diagrams comparing the typical H.A. solutions and what they solve and don't solve when used together or independently.


4. Is it recommended to deploy MSCS as "geocluster" splitting cluster nodes in separated sites (PROD & DR) with their owned shared storage given that bank has already a dark fiber interconnection between both sites (4 lines x 128 Gbps) using storage reploication (VPLEX)?
A question that is not MQ specific. The MSCS experts in your company should easily answer this. If they can't, back away slowly and consider a different solution.


As I read your problem description, this seems like a great use case for the MQ Appliances, especially if you are going to use physical Windows servers (phew, I though something smelled funny in here)




------------------------------
Peter Potkay

Original Message:
Sent: Fri May 29, 2020 07:02 AM
From: NICK DAKORONIAS
Subject: MSCS Cluster deployment options for MQ High Availabilty

Hello MQ community users,

I have a bank customer who is curently using an old (obsolete) Wintel based MQ infra (MQ ver 5.3 to MQ v8.x) and wants to deploy a new MSCS cluster where all stand alone based queue managers from different domains will be hosted (after being migrated).

This old Wintel MQ infra serves 24 x 7 banking applications and connects with queue managers on other platforms (AIX, z/Os).
The reason for maintaining the old MQ infra up to now, was the lack of MQ skills & expertise.

Customer's main scope for the deployment of MSCS MQ HA cluster is to minimize MQ apps downtime and to have a safe patching window (as part of security audit process), which is not currently feasible.

The solution options considered are either MSCS cluster Active/Passive (A-P) or Active/Active (A-A).

I would like to ask the following:
1. Which is the preferable MSCS cluster option (A-P or A-A) to be deployed in terms of MQ Service, objects & meessage availablity?

2. Is it recommended to deploy both MSCS cluster nodes as VMs (virtualized) instead of physical?

3. Is it feasible to combine MQCluster (shared cluster queues on multiple queue managers) with MSCS Active/Passive or Active/Active? If yes, which are the pros and cons?

4. Is it recommended to deploy MSCS as "geocluster" splitting cluster nodes in separated sites (PROD & DR) with their owned shared storage given that bank has already a dark fiber interconnection between both sites (4 lines x 128 Gbps) using storage reploication (VPLEX)?

P.S.
I have reviewed relevant IBM Knowledge Center articles, but i need more customized responses for this particular case:

https://www.ibm.com/support/knowledgecenter/SSFKSJ_9.1.0/com.ibm.mq.pla.doc/q017161_.htm -> Design Clusters

https://www.ibm.com/support/knowledgecenter/SSFKSJ_9.1.0/com.ibm.mq.con.doc/q017820_.htm -> HA configurations

Any advise will be much appreciated.
Cheers Nick.

------------------------------
NICK DAKORONIAS
------------------------------