On Tue, 8 Mar 2011, H Chang wrote: > I have a question on how ceph maintains tcp connections. > > As I understand, given a cluster of N osds, ceph client can > potentially read from, and write to all N osds, if crush distributes > data randomly across all. Right. > Then does the client maintains persistent tcp connections to all osds? The kernel client closes out idle connections. The userspace implementation does not do that yet. > How about tcp connections among osds (for replication)? Does each osd > maintain persistent tcp connection to all the other osds, essentially > forming a clique? The OSDs maintain open connections to the nodes they peer with. This is related to the number of PGs they store. So in a 10,000 node cluster, if an osd has ~100 PGs, it will have ~100 peer connections (for 2x replication). For a 10 node system, each node will typically peer with every other node. sage -- 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
