Open bobbyt2012 opened 1 year ago
@bobbyt2012 As far as I know, the command buffer filling is simply indicative of the destination not keeping up with the plotter. In other words, it is waiting to write to the destination. I believe that the [PlotWriter] Waited x.x seconds for a Command to be available.
is an indication of IO wait time.
Indeed it is as @Jsewill mentioned.
The buffer is currently of fixed size. It will likely be swapped out and be configurable with a dynamic one during beta, however, you will still hit a wall at the end of a plot even with an increased buffer count if the destination is still slow as you will have to wait for the plot to complete writing. It will also use more RAM as you increase the buffer count for the plot writer.
Are there any new updates on configuration parameters related to the buffer counts? I'm seeing the 'Command Buffer Full' on some of my plotting even to a local NVME drive. Edit: Looking more into this, one of my NVME drives seems to be under-performing so it looks like that's an issue with that drive. I don't see this error on a different drive.
What happened?
Not sure what causes this error or if it qualifies as a bug, but I don't see any resource constraints, so I'm curious as to what causes the "command buffer" to fill.
Machine Specs:
2x 2690v2 576GB RAM Nvidia A4000 (16GB) Final Directory 6TB - (5x SAS 1.2TB in RAID 0)
Command Syntax:
./bladebit_cuda -c -f -n 6 cudaplot /mnt/sas/complete
Output example:
Step 2 completed step in 7.18 seconds. [PlotWriter] Waited 9.900000 seconds for a Command to be available. [PlotWriter] Command buffer full. Waiting for commands. [PlotWriter] Waited 11.045000 seconds for a Command to be available. [PlotWriter] Command buffer full. Waiting for commands. [PlotWriter] Waited 11.099000 seconds for a Command to be available. Step 3 completed step in 37.70 seconds.
Version
bladebit-cuda-v3.0.0-alpha1-ubuntu-x86-64.tar.gz
What platform are you using?
Linux
What ui mode are you using?
CLI
Relevant log output