[jboss-dev-forums] [Design of JBossCache] - Re: memcached client/server

[bela@jboss.com]

Yes, ExpirationPolicy would certainly be needed, compared to defining a policy for an entire region.

Regarding consisten hashing: the Data Partitioning feature calls for metadata, replicated to all JBC instances, which keeps track of allocation of stripes to instances.
However, as an alternative, as you mentioned, we could use consistent hashing, to define where data is placed in the cluster. This could also be used to derive the primary node and backup nodes for a given stripe, instead of using metadata.
Interesting comparison: Google File System (GFS) uses metedata, but it is maintained by a single central server. However, this is not an issue as it can be reconstructed by consulting the cluster nodes.
We have to think a bit more about whether to use consistent hashing, or metadata, to place data across a cluster. I suspect metadata gives us more flexibility in supporting more exotic setups. If we store local metadata in every node, then we don't have to replicate metadata across the cluster, but the (JGroups) coordinator could maintain it. When it goes down, the next-in-line takes over, soliciting metadata from all nodes...

View the original post : http://www.jboss.com/index.html?module=bb&op=viewtopic&p=4177298#4177298

Reply to the post : http://www.jboss.com/index.html?module=bb&op=posting&mode=reply&p=4177298