Hello.
I have a PVE box running a single LXC (ubuntu 22) on Notebook with a 2TB NVME SSD and a not so powerful i7-8550U CPU. I know that this isn't a typical server infrastructure, but I chose PVE instead of directly running a linux on the notebook for better manageability and better PBS backup options.
All 8 cores of the host CPU are assigned to the LXC.
During normal operation, the CPU usage (on the host, running the LXC ) is less than 20% , even during backups this gets not higher than 50%. IO delay is in the range of 10% during normal operation, and maxes out around 50% during the backup. I am also attaching a screenshot showing a 1 day max CPU/IO Delay chart, as well as a LOAD chart from the PVE GUI.
The PBS server is hosted on a different box, using 2x4TB HDDs, and there is a 1Gbps switched network between the boxes.This box does not seem to be the bottleneck.
The LXC contains approx 1TB of data (on NVME SSD), approx 10-20% of it changing on a daily basis.
The ISSUE:
The LXC does time-critical things, and when the scheduled backup starts, the process is not able to work properly - most probably due to CPU or IOWAIT issues.
I wasn't able to identify the exact cause.
For me it seems that the bottleneck is happening in the first 30 minutes of the backup (which normally lasts around 2 hours) - mainly at a stage when it matches the diffs/dirty flags.
My goal is to limit the impact of the PBS backup on the running LXC container, but have no idea how. I would be fine if the backup process lasted even twice longer, but should not have impact on the processes running on the LXC.
I tried to play with the vzdump.com settings on the host (ionice) - but noticed no difference.
Tried to adjust / limit the bandwidth on the PBS server - the backup took much longer, but the situation for the LXC process showed no improvement.
Any help would be greatly appreciated!
Thanks in advance.
PVE 7.3
PBS 3.0
I have a PVE box running a single LXC (ubuntu 22) on Notebook with a 2TB NVME SSD and a not so powerful i7-8550U CPU. I know that this isn't a typical server infrastructure, but I chose PVE instead of directly running a linux on the notebook for better manageability and better PBS backup options.
All 8 cores of the host CPU are assigned to the LXC.
During normal operation, the CPU usage (on the host, running the LXC ) is less than 20% , even during backups this gets not higher than 50%. IO delay is in the range of 10% during normal operation, and maxes out around 50% during the backup. I am also attaching a screenshot showing a 1 day max CPU/IO Delay chart, as well as a LOAD chart from the PVE GUI.
The PBS server is hosted on a different box, using 2x4TB HDDs, and there is a 1Gbps switched network between the boxes.This box does not seem to be the bottleneck.
The LXC contains approx 1TB of data (on NVME SSD), approx 10-20% of it changing on a daily basis.
The ISSUE:
The LXC does time-critical things, and when the scheduled backup starts, the process is not able to work properly - most probably due to CPU or IOWAIT issues.
I wasn't able to identify the exact cause.
For me it seems that the bottleneck is happening in the first 30 minutes of the backup (which normally lasts around 2 hours) - mainly at a stage when it matches the diffs/dirty flags.
My goal is to limit the impact of the PBS backup on the running LXC container, but have no idea how. I would be fine if the backup process lasted even twice longer, but should not have impact on the processes running on the LXC.
I tried to play with the vzdump.com settings on the host (ionice) - but noticed no difference.
Tried to adjust / limit the bandwidth on the PBS server - the backup took much longer, but the situation for the LXC process showed no improvement.
Any help would be greatly appreciated!
Thanks in advance.
PVE 7.3
PBS 3.0
Attachments
Last edited: