WebSphere capacity plan

[7280]

Hi,
I was requested to setup a websphere environment which should be up and running 24/7.
This will be a very critical production system, that will serve a website and a contact center.

I'm thinking about what version of websphere install (cluster ND or XD), what hardware buy and so on...

From my point of view it's important that, in case of software changes (update ears..), the system must be up and running. For example is it possible to upgrade only a node of the cluster and the switch the connections to the other one?

Please help me find documentation...

Thanks

 

[Einstein47]

There are a lot of things you will need to consider to be able to hit the goal of 24/7 - server maintenance, backups, and not just of your WebSphere application and hardware, but your database, LDAP server, not to mention the OS limitations. [teal](For example, we run on AIX and an older version WebSphere had a memory leak which caused the paging space on the server to get all used up REQUIRING a system reboot weekly. The latest version WebSphere v6.1 seems to have plugged the leak, so our current server has been up for 103 days. I still keep my fingers crossed.)[/teal]

It isn't a simple answer as far as what do to. And unfortunately the IBM web site is not geared to user centric questions like this, but instead highly technical sales pitches. Good luck there.

I wish I could offer better suggestions, but I've been doing WebSphere admin support for 6 years now, and I still haven't been able to navigate the maze and keep my system up 24/7. I can only hope your luck is better than mine.

Einstein47
“Evil abounds when good men do nothing.“ - Nelson R.
[[]Starbase47.com]

 

[chgwhat]

The best I have seen was 4 boxes, 2 for WebSphere Edge Server, that sprayed the traffic to 2 other servers, when we had planned or unplanned activities, we would direct traffic to the other box, and by using a sprayer for the dns and another ip for testing, we could do a release, verify it on prod before it was available to the public, then just switch traffic to the new release and update the second box, the only outages we had were updates to a single backend component, like a database. We had cluster for the backend, which protected from hardware and os updates, but database updates always seemed to be the ones we had outages for.

Tony ... aka chgwhat

When in doubt,,, Power out...

 

[7280]

Thanks for your responses.

So it doesn't seem to be so simple!
Don't worry about database and OS backup, my big problem is to be able to update software (ears...) and to let end user work.
This is the big issue.

I was thinking about two physical nodes with a load balancer in front. During normal work, the traffic is send to both nodes. When I need to update a node, the load balancer will direct the traffic to the other node. Once the update is finished I'll do the same thing on the first node.
Does this make sense for you?

Thanks again.

 

[chgwhat]

That makes sense to me. Then you basically can have a simple websphere setup. And if you need a more capacity you could put 2 nodes on each server.

Tony ... aka chgwhat

When in doubt,,, Power out...

 

[Diancecht]

Just a thought: 24/7 doesn't mean you can ocasionally do a reboot.

Cheers,
Dian

 

[7280]

No reboot!

I was studying the Extendend Deployment cluster, it seems the right way to go.
Have you ever worked with? Is it stable?
IBM states that it's a consolidated version and that works perfectly.

Thanks again

 

[ntenz]

From my point of view it's important that, in case of software changes (update ears..), the system must be up and running. For example is it possible to upgrade only a node of the cluster and the switch the connections to the other one?"

We are in the same situation - Are there any guides to setup a cluster, scalable websphere application server enviroment?.

 

[7280]

WebSphere XD is the solution.

Or two identical clusters, WebSphere ND, in which you do manually what XD does for you.

I want to be sure that XD is consolidated and that it really does what's supposed to do!

 

[ntenz]

There have to be guides or other documentation regarding these setups. We need a 4 cluster system on 2 nodes. And need deploy new releases on 1 instance - verify and deploy on the 3 other systems. I will do some research regarding XD. Any links or other relevent issues regarding this subject?

 

[7280]

http://www.redbooks.ibm.com