Large number of cache instances

[dror76]

Hi

We are currently examining various ways of using Infinispan for a large scale multi-chat application .

One of the ideas is to create a cache instance per chat room - is means that at peak time we will have tens of thousand cache instances.

Our setup contains a varying number of servers on EC2 with Infinispan in replicated cluster mode (async).

We are looking for someone who already experienced this kind of usage or has a banchmark showing that Infinispan can handle this amount of cache instances.

Thanks,

Dror

[manik]

We haven't tried this with tens of thousands of cache instances, but I don't conceptually see anything wrong with it provided you have enough memory and horsepower on each node.

Quick question though, why not just use one cache for all of this, and "namespace" each entry by using "ChatroomName_<key>" as your key?

[ckulenk]

The DeltaAware/Delta interfaces may be of interest too. DeltaAware for a Chatroom, Delta for a message...

[navs]

Hey

I have a question on the same topic

When a map of objects is maintained in the cache, and a object changes, is the whole object resent for replication or only the dellta part?

The object are simple once with a few fields in it....

[manik]

The whole object.  Unless your map is an AtomicMap, created using the AtomicMapLookup helper.

[navs]

Hey Manik

Thanks for the reply!

I have one more question, If I use a nested AtomicMap will only the parent map be delta replicated or even the nested AtomicMaps will be delta replicated?

[navs]

another question, which is advisable a cache with structure <String,HashMap<String,SmallObject>>  or hundreds of caches, one for each SmallObject (keeping frequent changes to SmallObject and need to delta replicate them in mind).

Is there a big overhead in having many caches vs 1 cache?

[ckulenk]

Why not <KeyPath,SmallObject> with KeyPath={key1,key2,...} or something similar? I think massive amounts of caches aren't a good idea.

[manik]

AtomicMaps weren't designed to be nested.  I think this will cause issues.

[manik]

Yes, this is certainly a better option.

[navs]

Hey Manik

Thanks for you replies!

I am trying to implement a DeltaAware SortedSet (call it DeltaAwareTreeSet) for my application. Is is just enough to implement DeltaAware and Delta interface, and externalize delta object (call is TreeSetDelta)

I see AtomicHashMap has its own implementation of externalizer which is registered in Eternalizer class and mapped to it by ID assigned to AtomicHashMap.

If I create my delta aware SortedSet, how would I implement externalizer for my class?

[navs]

hey....the imlementaion worked fine!...never mind