> With regards to robustness, you state that there is a backup of
> each entry in the cache. How is this maintained? Is durability
> when writing to Coherence the same as writing to a typical
> relational database? I.e. when my transaction synchronously
> commits, it is there regardless of subsequent failure?
> Or can you only loose x nodes (partition distribution), etc?

Writes to the cache are backed up synchronously to another node in the cluster. Upon the loss of a node (or nodes) the backup copies will become primary copies, and these nodes will create their own backups. Note that the backups for a node will be spread across the cluster; there won’t be a single backup node for a node.

This document provides a good description of the data partitioning and goes into a bit more detail:

http://coherence.oracle.com/display/COH34UG/Partitioned+Cache+Service

> Can you point to some resources on how the cache works? I.e. does it
> partition the datastore across the nodes? Does it cache via LRU on
> each node? When you write to the cache how to make the guarentee that
> each node sees a consistent copy?

As the data is partitioned across nodes, each node will be responsible for persisting to the data store its share of data. It also means that for a given key, only one node at a time is considered the Source of Truth. This is why the entire cluster sees a consistent data set; everyone is going to the same node to read/write that value.

The eviction policy is configurable; by default it uses a hybrid LFU/LRU algorithm.

This document goes into more detail about database integration:
http://coherence.oracle.com/display/COH34UG/Read-Through%2C+Write-Through%2C+Write-Behind+and+Refresh-Ahead+Caching

Thanks for the comments!
-Patrick

Can you point to some resources on how the cache works? I.e. does it partition the datastore across the nodes? Does it cache via LRU on each node? When you write to the cache how to make the guarentee that each node sees a consistent copy?

Thanks so much-
Steve