On Wed, Aug 29, 2012 at 12:46 AM, hemant surale <[email protected]> wrote: > Hi Tommi , > Actually I was thinking of "Availability" of related files to > particular host. I wanted to guide ceph in some way to store related > files for host on his local osd so that I/O over network can be > reduced .
That sort of a design is only possible if you have a lookup table for data locations, like e.g. HDFS does. Which has it's own problems, causing a lot of lookup traffic and contention. CRUSH means data placement does not depend on the origin of the data, only on the object name and CRUSH ruleset. Frankly, if you're thinking about single machines, you're probably thinking at a smaller scale than what Ceph is aimed at. > One more doubt I would like to ask is Whether read/write > requests are sent to primary only Or only write requests are sent to > primary while read request can be served by any OSD having object > replica ? Writes are sent to the primary, which sends them on to all the replicas. Read requests are served by the primary also. There has been some interest in reading from any replica, especially in the Hadoop world, but as far as I remember that feature is not implemented. -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html
