Hello, On Wed, 15 Apr 2015 10:26:37 +0200 Atze de Vries wrote: > Hi, > > We are designing a new Ceph cluster. Some of the cluster wil be used to > run vms and most of it wil be used for file storage and object storage. > We want to separate the workload for vms (high IO / small block) from the > bulk storage (big block lots of latency) since mixing IO seems to be a > bad idea. > I assume different pools with differently backed OSDs then? > The new NVMe disks (for example Intel P3600 or P3700) can handle 64K > queues with 64K commands each while AHCI can handle 1 queue with 32 > commands. > NVMe indeed addresses that shortcoming of AHCI nicely. > Would the performance loss from mixing IO be less if you would use NVMe > disks? > I encourage you to read all the various SSD cluster performance threads here. In short, you're likely to run out of CPU for IOPS long before scratching the IOPS potential of these NVMe SSDs. Christian -- Christian Balzer Network/Systems Engineer chibi@xxxxxxx Global OnLine Japan/Fusion Communications http://www.gol.com/ _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |