On Tue, Jul 22, 2014 at 2:48 PM, Shayan Saeed <shayansaeed93@xxxxxxxxx> wrote: > Another question along the same lines. For erasure code, same as > replicated files, the request goes through the primary member. Isn't > it possible to send the request to any of the members and get the > file. While this might have kept things neater on the development side > and might have made some sense for replicated system, it makes the > availability and load balancing worse for erasure coded files. I see a > lot of requests coming in for a specific object which makes the > primary osd hosting it go down sometimes and then all the requests > have to wait another osd comes up and repair is done. For load > balancing purposes, is there a way to make the requests go to someone > else without hinderance and get the object without waiting for repair. Not really. This would have all the same consistency issues of having replicated objects read from multiple replicas, plus the problem of needing to get a consistent view across multiple machines instead of a single machine. If an OSD is getting overloaded by clients, that's the bug and it should be addressed as one. -Greg Software Engineer #42 @ http://inktank.com | http://ceph.com -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
