I went ahead and pinned 2 cpus and had a cpu stall and crash.  I've seen stalls before in the log but no crash.  So I went back to all 8 cpus and 4 gb.

I went ahead and pinned 2 cpus and had a cpu stall and crash.  I've seen stalls before in the log but no crash.  So I went back to all 8 cpus and 4 gb.

 

quick question, did you also exclude the pinned cores from your domU's? if not this will actually make the cpu starvation issue potentially worse, as you then only have 2 cores allocated to dom0 AND they are also still being used by your domU's.

 

you need to tweak your cfg files for each domU as follows, this is an example and assumes you have allocated core 0 and 1 to dom0, and your cpu has at least 6 cores:-

 

#number of cores allocated to to this vm
vcpus = '4'
cpus = '2-5'

 

so this will allocate core 2 to 5 for this particular vm, which is obviously 4 cores, the name of the game is do not allow any vm to use your pinned cores for dom0, thus guaranteeing that xen running in dom0 does not get cpu starvation due to high cpu usage in your domU's.

 

for me this has completely stabilised my system, i have an uptime now of more than 2 weeks, previously to this i was having crashes every 4-5 days.

 

just incase anybody wants to look at how to pin cores for dom0, here is my boot option from my syslinux.cfg:-

 

label Xen/unRAID OS (experimental pinned core)
  menu default
  kernel /syslinux/mboot.c32
  append /xen dom0_max_vcpus=2 dom0_vcpus_pin --- /bzimage --- /bzroot

 

im assuming some/most people aren't seeing the crash issue due to either their cpu being so powerful that it can handle processing both dom0 and domU at the same time under full load, or just the simple fact that people arent running cpu intensive tasks in their vm's and thus dom0 has plenty of cpu cycles free to process xen commands.

 

i would be interested to hear from people running say multiple plex transcodes and not having anything pinned and still seeing no issues, i wonder if the issue is more prevalent in AMD processors, hmmmm.

No I didn't add the cpu line to the vm cfg. I only use 1 vm 24/7 &  I have 8 cores so I will add 4-7.

 

I never had a stability problem before and can run unpinned without issue.  My goal was to isolate my vm even more from possibly using the same cpu. During parity checks I still get a little bit of artifacts and buffering of mythtv backend.  I have my vm drives on its own controller which helps. It just might not be possible with this hardware.