[ Adding ceph-users to the CC ] On Mon, May 11, 2015 at 8:22 PM, zhao.mingyue@xxxxxxx <zhao.mingyue@xxxxxxx> wrote: > Hi： > > I'm learning CephFS recently, and now I have some question about it; > > > > 1. I've seen the typical configuration is 'single MDS',and found some > resources from Internet which said 'single MDS' is stable; > > > > If lots of I/O requests send to the only MDS, performance of the whole > filesystem would be not very well, isn't it? Yes, the metadata throughput of a single MDS is a potential limiting factor right now. Keep in mind that the CephFS protocol is very different from many other distributed filesystems so this is likely to be less of a problem than it is for others — clients can satisfy *many* operations from their local cache so it's usually only on first access or flushes that a metadata request has to go to the MDS. > 2. Compared to 'single MDS',' several MDS' maybe is not stable enough > ------can you give me a detailed explaination? such as :according to what > come to this conclusion,what is the unstable element…… A multi-MDS system has to do a lot of coordination work between the MDSes that a single-MDS system does not. These code paths are less tested and less stable than the ones between clients and the single MDS. -Greg _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |