On Apr 17, 2012, at 9:36 AM, Sage Weil wrote: > I guess the question is, is there a compelling use case, and if there is, > is there a better approach?
I am failing to find a reference to the paper, but I recall that the 1 localized write to HDFS for MR applications isn't enormously beneficial. Then again it does save a lot of network traffic, especially for triplication. As more people start looking at the Object Class handler mechanism, I could see it being desirable to stick data into objects (grab later via rendezvous) rather than return data to the rados_exec caller. Making that efficient would be important. Could you piggyback on the recovery logic to accomplish this? Say, write 2 objects as normal, and 1 local "miss placed" object, letting RADOS asynchronously move that object to its intended location. - Noah-- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
