5 Replies Latest reply on Nov 3, 2011 5:12 AM by galder.zamarreno

using infinispan for replicating data (not just cache)

dex80526 Oct 26, 2011 10:08 PM

I am looking for a solution for following use scenario:

1) a cluster of nodes share common data (more reads and few writes).

2) the cluster needs to be able to scale up or down by adding/removing nodes dynamically.

3) right now, the data is stored on external shared storage.

For the scability and other consideration, I'd like to get rid of the shared storage, and move to a "shared nothing" cluster.

After read some articles/documents on Infinispan, conceptually, I feel that Infinispan may fit the use case. I am thinking to use infinispan in a cluster mode (with replication) and "write through". I could use the "distributed" mode here.

But, I did not find any real example how Infinispan's "write through" is implemented.

Anyway, does Infinispan support this kind of use cases? Does anyone have any experience with Infinispan for this kind of use scenario?

I'd like to here any suggestions.

Thanks

1. Re: using infinispan for replicating data (not just cache)

galder.zamarreno Oct 31, 2011 12:19 PM (in response to dex80526)

Write through? You mean you want a cache plugged with a cache store, and write through to it?

The rest of bits look sensible, but it's not clear how you're planning to access Infinispan, whether in embedded (in-memory) or remote way.

There's plenty of info in https://docs.jboss.org/author/display/ISPN/User+Guide, particularly around the server module pages.
Actions
2. Re: using infinispan for replicating data (not just cache)

dex80526 Oct 31, 2011 6:02 PM (in response to galder.zamarreno)

Thanks for the respsone.

The ISPN is going to be embeded.

The whole idea is to have the data be replicated/distributed among the nodes. The cache store (data store) is on each node. I.e., there is no common data store (such as a database behind, which is we want to get rid of).

Other types of implementation I looked at such as XtremeFS (http://www.xtreemfs.org/). I feel we could achieve what XtremeFS with ISPN at the application level instead of FileSystem level.
Actions
3. Re: using infinispan for replicating data (not just cache)

galder.zamarreno Nov 1, 2011 1:24 PM (in response to dex80526)

Right, so if you're not using a cache store, I don't understand what you mean by write-through, because in caching terms, that's normally used to talk about a backing cache store, and how you write through it.

If you're talking about write-through to other nodes in the cluster, see the section on cache modes in the guide.
Actions
4. Re: using infinispan for replicating data (not just cache)

dex80526 Nov 1, 2011 3:01 PM (in response to galder.zamarreno)

In terms of cache store, I am thinking using FileSystem cache store, in this case.

Yes, I'd like to have any changes to the cache written out to FileSystem on each node.

It's not clear to me how to handle data sync when a new node joins the cluster, or a node craches and rejoins the cluster.

I am going to re-read those sections on cache mode. Thanks.
Actions
5. Re: using infinispan for replicating data (not just cache)

galder.zamarreno Nov 3, 2011 5:12 AM (in response to dex80526)

When a node joins the cluster, state transfer can be used so that new joining node can retrieve all data on startup from other nodes in the cluster. If you use sync replication, it's enabled by default. Alternatively, you can disable state transfer and instead use a cluster cache loader to retrieve data lazily.
Actions

Go to original post