Hi,

I have an environment with 2 cells and one websphere cluster in each cell. I've deployed the same application in both clusters in order to reach high avaibalility and desployments with no downtime.

My application uses a websphere object cache instance. ¿Is there any way to sinchronize both object cache instances (one per cell) in order to avoid consistency data problems?
 
Than you very much! 

The options available are good to replicate the objectcache with in one cell and one replication domain but i don't think they can do between multiple cells. 


just a curious question : Is your two cell/cluster architecture for same application,  a failover model or load balancing model ? 

It's just a proof of concept. I'm thinking about a was nd topology with no downtime in deployments. 

I've read this topology recomendation in IBM high availability guide for WAS 6. 

As you say, persistent resources like sessions, messages and transactions are not failovered beetween cells, you may wait for them to finish before stopping a cell (branch). 

But, what happens with caching objects when both cells are running? If there is no sync between cells, it involves a data consistency problem. Doesn't it? 

I actually have never seen/worked on this approach. 

If you take a cell as an app domain ... will you have two of them working at the same time? In general we go for a backup/failover , which can pickup in case of a downtime. This also we do as network level. also did you considered creating another cluster on different nodes in the same cell to deploy the same app ? Then you can make use of the replication domains. 

As for the loadbalancing... it can be achieved in a single cell with multiple cluster members on multiple nodes.  

If the downtimes for deployments are the big constraint .. checkout the option "rollout update".  

This is a totally different approach to addressing your core need of minimizing downtime, so I appologize now if it's a distraction...

If your end goal is to reduce the overall downtime for the appliction, can I suggest you take a look at the WebSphere Virtual Enterprise (WVE) product if you haven't already?

This is a high end version of the WebSphere Application Server Network Deployment.  It has a wealth of features, but here are a few related to minimizing downtime:
1. Allows for multiple versions of your application code to be deployed concurrently;
2. Priority routing of requests through intelligent routing at the web server;
3. Proavtice administration tasks to handle conditions you can define in the JVM (think memory leak for example;)
4. SLA support
5. Dynamic clusters (brining up and shutting down JVMs as demand varies.)

As someone whose investigated the WVE product, and even been through the full Proof of Technology with IBM on site, I think it might be what you're looking for.

While each environment varies, WVE might actually be cheaper in the long run compared to building an entirely new ND cluster (depending on your topology.)

On a final note, the technology that performs the magic of WVE is also included in the IBM Workload Deployer product...  That's a whole need beast that can do some great provision very quickly! 

Sorry if I sidetracked your thread.  Hopefully there's some value in it for you.  If you have additional questions, you might want to post a new thread asking about WVE rather than use this thread.

Erik

check this developerworks article: www.ibm.com/developerworks/websphere/tec...

Tom Alcott is one of the best WebSphere experts out there on HA & DR in my opinion.  If you ever get a chance to see him at the IMPACT conference, he does a great job presenting.

That said, I think you bring up a great point Joseph.  What degree of HA or DR do you need?  Alternatively, what amount of downtime is your employer willing to accept?  Everything costs money, and each 9 that you add to you 99.9% can cost you  exponentially more...

Erik 

yes ... given the technology and options, we can achive the 9's but always the constraint is budget :-)