If you are unable to create a new account, please email support@bspsoftware.com

 

News:

MetaManager - Administrative Tools for IBM Cognos
Pricing starting at $2,100
Download Now    Learn More

Main Menu

Load Balancing Dispatchers

Started by cpasquini, 21 Sep 2010 09:30:50 AM

Previous topic - Next topic

cpasquini

Hi All,

I am a first time poster to this forum.  I am running into an issue with my cognos environment (I support it, but do not code or create reports) that I believe is caused by the dispatcher setup.  Here is a quick overview of the setup:

1.  Application Front-end with ServletGateway webapp which connects to the dispatcher via hostname:port
2.  Dispatcher service with content store

Our dispatcher service is actually running on two linux servers (with identical specs) for redundancy purposes.  We are using hardware load balancing (via an 'F5' network device) on top of the dispatchers.  The idea is that the ServletGateway knows only about the dispatchers via a hostname (i.e. 'cognos.service.com:9003').  The F5 device uses a virtual IP setup, meaning the hostname resolves to an IP that has a pool of 2 servers - the 2 server being my Cognos dispatchers.  We set up the load balancing so that server 1 takes all of the requests.  In the even that server 1 goes down, server 2 will take the traffic. 

Within Cognos administration however I always see two dispatchers available - I know that this is because each dispatcher connects to the same content store (which we want).  I do not believe that our load balancing method is working as expected since I would expect to never see traffic in server 2 - unfortunately I do.

Our goal is to setup one dispatcher to handle 100% of the calls and use the second dispatcher has a 'hot standby'.  In the event that dispatcher 1 goes down or becomes unavailable we would like all subsequent calls to be handled by dispatcher 2.  At no point do we want to send requests to server 2 while server 1 is up and running.

The reason I believe that this is a problem is because it seems that when two users are in the application at the same time generating reports, one report will complete without a problem, but the other report will eventually time out after four minutes, throwing an DPR-ERR-2002 error in my dispatcher 2 server log. 

Has anyone setup their dispatcher's in this manner with load balancing?  I am considering activating the Cluster Compatible load balancing mode in cognos admin, but the documentation available is not very good, nor do I know if it will solve this issue.

Thanks in advance!

Chris 


RobsWalker68

Hi Chris,

If you have 2 dispatchers sharing a content store then you will correctly see both dispatchers in Cognos Admin.  The dispatcher on startup will register itself with the content store.

If you are using hardware load balancing and you do not want to use the default round robin load balancing then you need to switch to Cluster Compatible.  At the moment the hardware load balancing may be directing requests to server 1 but Cognos will alos be software load balancing.

Kind Regards

Rob   

technomorph

Hi Chris,

I'm not sure if you're using a different gateway to the C8 standard, but is there much value in having a dedicated fail-over dispatcher? The default configuration using round-robin load balancing will ensure requests are serviced as efficiently as possible (using both servers) whilst at the same time providing fail-over by allowing one of the servers to pick up all requests if the other fails .

You're current set-up will provide fail-over support, but the concurrency capacity is half what it could be no??

Cheers

cpasquini

Thanks for the replies everyone.  :D

So is switching to cluster compatible as simple as just editing the load balancing mode in Cognos Admin->System Settings?  If we make the change is it immediate or do we need to make any additional configuration changes (such as setting up server groups etc, bouncing the dispatchers etc)?

technomorph - it's more of a licensing issue - we are only licensed to run the Bognos BI on a single machine - the hot standby is meant to address the issue of redundancy, not necessarily capacity.  We may need to add capacity at a later time as the application takes on more users.

Chris


RobsWalker68

Hi Chris,

From what I can remember making the change to cluster compatible in the admin consol should work through without the need to restart the services.

Another way to approach this would to be using dispatcher routing.  You can based on packages/users/groups determine which dispatcher process what requests.  Obviously, this is a bit fiddlier should you have to switch to the "passive" dispatcher

Regards

Rob

cpasquini

OK - I think we are going to implement the cluster compatible LB mode this weekend.  As an update to the original issue of some reports failing with the DPR-ERR-2002 error, I shutdown my hot-backup dispatcher and restarted it.  Once I did that the issue went away.  Some somehow reports were getting passed to that hot-backup and the restart cleared that out.

Chris

FM