| Author |
Message
|
| queuemanager |
 Posted: Sun Sep 13, 2009 10:52 pm Post subject: Number of full repositories |
 |
|
Apprentice
Joined: 28 Nov 2006
Posts: 43
Location: Bangalore
|
Hi,
I have a question about number of full repositories in a cluster, I know the manual says that we should have 2 full repositories in a cluster and having more than two should be suported by a very good reason. Now two questions :
1. How often full repositories send messages to each other to be in sync?
2. How much of a performance impact is there if I configure 4 full repositories in a 15 queue manager cluster rather than 2?
We are in a process of finalizing our design for cluster topology for our environment and we are planning to have cluster with queue managers in NA, EMEA and APAC.
Any help is appreciated.
Thanks |
|
| Back to top |
|
 |
| exerk |
| Posted: Mon Sep 14, 2009 12:08 am Post subject: |
|
|
Jedi Council
Joined: 02 Nov 2006
Posts: 6339
|
Read THIS, hopefully it will help you in your decision making. If your number of 15 queue managers in the cluster includes the proposed 4 FR's, nearly one third of your queue managers will be FR's, and in such a small cluster I can see absolutely no justification for having more than two.
_________________
It's puzzling, I don't think I've ever seen anything quite like this before...and it's hard to soar like an eagle when you're surrounded by turkeys. |
|
| Back to top |
|
|
| queuemanager |
| Posted: Mon Sep 14, 2009 12:35 am Post subject: |
|
|
Apprentice
Joined: 28 Nov 2006
Posts: 43
Location: Bangalore
|
Exerk, thanks for your response.
Suppose I have 8 QMGRs in NA and 7 in EMEA, so should I go with a configuration where I have 1 FR each in both the regions?
Also I was curious about how often the FRs exchange information other than when there is a change in cluster object definintion or a new definition? and what kind of information is exchanged, is it the full info in the FR that flows or is there some kind of checksum technique?
If the information flows once every 2-3 hrs than having two full repositories across atlantic can be considered, but if it flows every 5 mins then this could be an issue.
Thanks |
|
| Back to top |
|
|
| queuemanager |
| Posted: Mon Sep 14, 2009 12:40 am Post subject: |
|
|
Apprentice
Joined: 28 Nov 2006
Posts: 43
Location: Bangalore
|
Also, the reasons given in the blog suggest that if suppose only little information flows to keep FRs in sync and that too after 2-3 hrs then the only significant disadvantage would be administering higher number of FRs when changing the set of full repositories or at the time of migration.
Appreciate your views.
Thanks |
|
| Back to top |
|
|
| exerk |
| Posted: Mon Sep 14, 2009 12:46 am Post subject: |
|
|
Jedi Council
Joined: 02 Nov 2006
Posts: 6339
|
| queuemanager wrote: |
| ...Suppose I have 8 QMGRs in NA and 7 in EMEA, so should I go with a configuration where I have 1 FR each in both the regions? |
Sounds reasonable. Make sure they are the most stable and available boxes you have in each area.
| queuemanager wrote: |
| Also I was curious about how often the FRs exchange information other than when there is a change in cluster object definintion or a new definition? and what kind of information is exchanged, is it the full info in the FR that flows or is there some kind of checksum technique? |
The honest answer is "I don't know the detail", but apart from when a cluster is first set up, or when a refresh/reset cluster command is issued, I don't think a lot of cluster management-specific traffic flows during 'normal' operations. Of course, I could be wrong, and am happy to be put straight.
| queuemanager wrote: |
| If the information flows once every 2-3 hrs than having two full repositories across atlantic can be considered, but if it flows every 5 mins then this could be an issue. |
And every 5 minutes wouldn't be an issue with 4 FR's set in a 'ring', as per the link I posted? 
_________________
It's puzzling, I don't think I've ever seen anything quite like this before...and it's hard to soar like an eagle when you're surrounded by turkeys. |
|
| Back to top |
|
|
| mvic |
| Posted: Mon Sep 14, 2009 1:26 am Post subject: |
|
|
Jedi
Joined: 09 Mar 2004
Posts: 2080
|
| queuemanager wrote: |
Also I was curious about how often the FRs exchange information other than when there is a change in cluster object definintion or a new definition? and what kind of information is exchanged, is it the full info in the FR that flows or is there some kind of checksum technique?
If the information flows once every 2-3 hrs than having two full repositories across atlantic can be considered, but if it flows every 5 mins then this could be an issue. |
Unlikely you will ever see traffic between the two full repositories except when qmgrs in the cluster are creating or changing their clustered objects. If info ever flows it is on a per-object basis, not the whole repository. |
|
| Back to top |
|
|
| queuemanager |
| Posted: Mon Sep 14, 2009 1:30 am Post subject: |
|
|
Apprentice
Joined: 28 Nov 2006
Posts: 43
Location: Bangalore
|
| Quote: |
| And every 5 minutes wouldn't be an issue with 4 FR's set in a 'ring', as per the link I posted? |
Every 5 minutes could be an issue but every 2-3 hrs may not be with repect to performance overhead, thats when there is minimal information is flowing across the cross atlantic link. |
|
| Back to top |
|
|
| exerk |
| Posted: Mon Sep 14, 2009 1:42 am Post subject: |
|
|
Jedi Council
Joined: 02 Nov 2006
Posts: 6339
|
| queuemanager wrote: |
| Quote: |
| And every 5 minutes wouldn't be an issue with 4 FR's set in a 'ring', as per the link I posted? |
Every 5 minutes could be an issue but every 2-3 hrs may not be with repect to performance overhead, thats when there is minimal information is flowing across the cross atlantic link. |
I don't think you understood my query...so I'll restate it:
If 2 FR queue managers updating each other every 5 minutes is a concern in terms of network load/performance etc., how much worse is it going to be with 4 FR queue managers updating each other every 5 minutes?
_________________
It's puzzling, I don't think I've ever seen anything quite like this before...and it's hard to soar like an eagle when you're surrounded by turkeys. |
|
| Back to top |
|
|
| sumit |
| Posted: Mon Sep 14, 2009 1:45 am Post subject: |
|
|
Partisan
Joined: 19 Jan 2006
Posts: 398
|
Most of the times, FR communicates when there is some change at PR level. In a well developed env, there are not much msgs flow between 2 FRs.As you are in initial stage of development, the frequency of communication can be higher which will eventually go down as you reach stability.
I think number of FR also depends on the outage period in various locations. There should not be a case when you have outage at the same time on both the places where you have FR. Would this be the case, the count of full repos may increase.. i.e may be from 2 to 3.
_________________
Regards
Sumit |
|
| Back to top |
|
|
| queuemanager |
| Posted: Mon Sep 14, 2009 2:39 am Post subject: |
|
|
Apprentice
Joined: 28 Nov 2006
Posts: 43
Location: Bangalore
|
Sorry I misunderstood, The reason I wanted to know about the overhead is that we have another design in mind wherein we are planning to have 2 different clusters for each region and making queue managers PRs for other cluster if they host any object that needs access across region. If the performance overhead is not significant, and not much of information flows among FRs then we may have two full repositories each in both the regions as high availability of full repositories is also one of the major consideration we have.
| Quote: |
| Most of the times, FR communicates when there is some change at PR level. |
Not sure.....I'm not sure but I think there has to be some mechanism to ensure that FRs are always in sync. |
|
| Back to top |
|
|
| exerk |
| Posted: Mon Sep 14, 2009 2:42 am Post subject: |
|
|
Jedi Council
Joined: 02 Nov 2006
Posts: 6339
|
| queuemanager wrote: |
| ...we are planning to have 2 different clusters for each region and making queue managers PRs for other cluster if they host any object that needs access across region... |
Overlapping clusters would achieve that.
_________________
It's puzzling, I don't think I've ever seen anything quite like this before...and it's hard to soar like an eagle when you're surrounded by turkeys. |
|
| Back to top |
|
|
| mvic |
| Posted: Mon Sep 14, 2009 3:31 am Post subject: |
|
|
Jedi
Joined: 09 Mar 2004
Posts: 2080
|
| queuemanager wrote: |
| Not sure.....I'm not sure but I think there has to be some mechanism to ensure that FRs are always in sync. |
Yes, whenever one FR hears of an update to an object, it forwards the update to its partner FR. |
|
| Back to top |
|
|
| PeterPotkay |
| Posted: Mon Sep 14, 2009 5:28 am Post subject: |
|
|
Poobah
Joined: 15 May 2001
Posts: 7722
|
|
| Back to top |
|
|
| JosephGramig |
| Posted: Mon Sep 14, 2009 12:19 pm Post subject: |
|
|
Grand Master
Joined: 09 Feb 2006
Posts: 1244
Location: Gold Coast of Florida, USA
|
In the original post, it was indicated that QMGRs exist in two distinct regions that have little interaction and a small bandwidth between them.
Having two clusters that are bridged probably makes the most sense. Do not use the FRs as the bridge QMGRs.
In fact, if you value your cluster, you will isolate your FRs from any other activity. Isolation will protect the FRs from other activities like "REFRESH CLUSTER" or suspending the QMGR from the cluster to do some work.
Finally, having isolated FRs allows you to migrate them first and apply maintenance to them first. A good idea.
If you are so poor that you must run the FR on a machine that does other stuff, then create another QMGR that is just the FR. Never allow anybody connect directly to the FRs via Client channels. |
|
| Back to top |
|
|
|
|
