Ceph:Best practices
Jump to navigation
Jump to search
Best Practices
Plan it!!! Requirements: size, performance, budget
Cost - vs reliability node failures Rack failure? Multi site replication Plan for redundancy of monitors nodes 3 copies = 8 nides of availablity = less than 1 second of downtime per year.... Everything effects performance - in above and below ceph Understand your workload
NUMA and NUNA can effect performance Multi site - can cause latency issues, applications could time out...
Recommended Storage server config
CPU - 1 core GHz per OSD Memory - 2GB per OSD Disk Controller - no extender for data and journal, RAID1 for OS disks Data - size doesn't matter but 12 disks per OSD is usually optimum Journal Disks - SSDs, write optimized
NUMA
Populate all channels to maximise bandwidth Use sensible memory configs