Hi,
We have a 25 node Proxmox cluster using Ceph as the storage with 200 Samsung PM883 1.92TB SSD drives as OSD's. We are troubleshooting an issue with lower than expected IOPS which isn't a major issue day to day but has caused us problems if Ceph needs to rebalance (eg drive swap).
While troubleshooting it has been discovered that write cache is turned on (hdparm -W /dev/XXX). I suspect this used to be disabled but has changed since the nodes got rebooted a while ago.
1. From reading around it seems to be recommended that write cache is disabled across all drives - is this still the recommendation?
2. Is there a way of setting this so it continues to apply after reboots?
3. Is it safe to disable cache on all drives (hdparm -W 0 /dev/XXX) within a live cluster outside of a maintenance window, ie is there any chance of service interruption?
Thanks
We have a 25 node Proxmox cluster using Ceph as the storage with 200 Samsung PM883 1.92TB SSD drives as OSD's. We are troubleshooting an issue with lower than expected IOPS which isn't a major issue day to day but has caused us problems if Ceph needs to rebalance (eg drive swap).
While troubleshooting it has been discovered that write cache is turned on (hdparm -W /dev/XXX). I suspect this used to be disabled but has changed since the nodes got rebooted a while ago.
1. From reading around it seems to be recommended that write cache is disabled across all drives - is this still the recommendation?
2. Is there a way of setting this so it continues to apply after reboots?
3. Is it safe to disable cache on all drives (hdparm -W 0 /dev/XXX) within a live cluster outside of a maintenance window, ie is there any chance of service interruption?
Thanks