Hello everyone,
I'm running proxmox with lvm-thin on local hard-drive and I can't find any way to fix the huge iodelay on server when receiving/restoring/migrating a VM.
I've got this on Proxmox 5 and 6, with different servers generations : DELL R620 or R640 with PERC + Mechanical HD or perc + ssd (enterprise grade, not consumer), it's still the same. RAID1 or RAID5.
When using pveperf, it only check root (where is installed proxmox) harddrive. Do you know a way to bench disk local-lvm disk read/write ? And what should I have as a "minimum" value ?
I've modified the datacenter.cfg on proxmox6 to limit to 100Mo/sec which seems not to be gigantic to me on SSD, and I still have something between 5 to 15 in IOdelay. Thus, each vm inside the proxmox host hang or have huge slow down when trying to write on disk.
On proxmox6, I noticed at first glance it's when the disk image is zeroing. When copying image file, there is no iodelay. I've made a test on proxmox5, and it seems it's the same (I've less logs to tell but seems to be so...).
Can't we limit the bandwidth of the lvm-thin zeroing since it seems to be the culprit ? And not only the disk image copying (which is limited by cstream as I've seen) ?
Thanks for your help, I can't find anything of the forum that works and I guess i'm not alone (!?) :-(
Regards,
PS: Thanks to the proxmox team, you guys are awesome !
I'm running proxmox with lvm-thin on local hard-drive and I can't find any way to fix the huge iodelay on server when receiving/restoring/migrating a VM.
I've got this on Proxmox 5 and 6, with different servers generations : DELL R620 or R640 with PERC + Mechanical HD or perc + ssd (enterprise grade, not consumer), it's still the same. RAID1 or RAID5.
When using pveperf, it only check root (where is installed proxmox) harddrive. Do you know a way to bench disk local-lvm disk read/write ? And what should I have as a "minimum" value ?
I've modified the datacenter.cfg on proxmox6 to limit to 100Mo/sec which seems not to be gigantic to me on SSD, and I still have something between 5 to 15 in IOdelay. Thus, each vm inside the proxmox host hang or have huge slow down when trying to write on disk.
On proxmox6, I noticed at first glance it's when the disk image is zeroing. When copying image file, there is no iodelay. I've made a test on proxmox5, and it seems it's the same (I've less logs to tell but seems to be so...).
Can't we limit the bandwidth of the lvm-thin zeroing since it seems to be the culprit ? And not only the disk image copying (which is limited by cstream as I've seen) ?
Thanks for your help, I can't find anything of the forum that works and I guess i'm not alone (!?) :-(
Regards,
PS: Thanks to the proxmox team, you guys are awesome !
