Hi Mark, Thanks for your response, it is help! Our Ceph cluster use Samsung SSD 870 EVO all backed with NVME drive. 12 SSD drives to 2 NVMe drives per storage node. Each 4TB SSD backed 283G NVMe lvm partition as DB. Now cluster throughput only 300M write, and around 5K IOPS. I could see NVMe drive utilization over 95% show on ‘iostat’ command. Will NVMe drive be a bottle neck quickly if we have large of IO in cluster? I have read the top article about OSD bundle with CPU cores. However I can only find script called pincpu on the github to automate process to allocate CPU core with OSDs. It seems not work for me. Do you have any tool or official instruction that can guide me to test it? _______________________________________________ ceph-users mailing list -- ceph-users@xxxxxxx To unsubscribe send an email to ceph-users-leave@xxxxxxx
 |