> 7 авг. 2017 г., в 9:54, Wido den Hollander <w...@42on.com> написал(а): > > >> Op 3 augustus 2017 om 15:28 schreef Mike A <mike.almat...@gmail.com>: >> >> >> Hello >> >> Our goal it is make fast storage as possible. >> By now our configuration of 6 servers look like that: >> * 2 x CPU Intel Gold 6150 20 core 2.4Ghz >> * 2 x 16 Gb NVDIMM DDR4 DIMM >> * 6 x 16 Gb RAM DDR4 >> * 6 x Intel DC P4500 4Tb NVMe 2.5" >> * 2 x Mellanox ConnectX-4 EN Lx 25Gb dualport >> > > To get the maximum out of your NVMe you will need higher clocked CPUs. 3.5Ghz > or something. > > However, I'm still not convinced you will get the maximum out of your NVMe > with Ceph. > > Although you are looking into 'partitioning' your NVMe with SPDK I would look > at less core which are clocked higher. > > Wido >
Thanks for reply. The description of server have an error: not 20 core with 2.4Ghz per core, this is 18 core with 2.7Ghz per core. Yes, I also think that an any cpu that I can use will be not enough the maximum out of these disks. I think, maybe in future ceph will be better in CPU consumption issue and these CPU can server more IOPS than now. By now my idea is do decrease a CPU consumtion by use SPDK and RDMA technology. Hope that SPDK and RDMA is now prodaction use ready. >> What a status in ceph of RDMA, NVDIMM access using libpmem and SPDK software? >> How mature this technologes in Ceph? Ready for prodaction use? >> >> Mike >> >> — Mike, runs _______________________________________________ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com