On 13 Feb 2009, at 19:26, Justin Bertram wrote:
Manik,
You can't talk about "high-performance non-blocking state transfer" and not provide benchmarks! How high is "high-performance"?
I haven't mentioned any benchmarks since we don't have any as yet. The reason why it is "high performance" is that the design does not block concurrent cache activity while state transfer is taking place, while the older mechanism blocked not just the cache providing state, but the entire cluster from receiving updates. And if there was a lot of state to transfer (e.g., gigabytes) the cluster could be blocked for quite a while.
Benchmarks will come soon, we're working on that. :-) Feel like helping out?