This kernel works as expected with one exception. The exception has been a nagging problem, but I have not reported it because 1) we are using a research OS in DomU and 2) we are not clear if the problem is in our code, Linux or Xen. But, here are the symptoms:
Occasionally (this seems to correlate to network activity between Dom0 and DomU), the system becomes unresponsive. I am running the Michael Young kernel at runlevel 3 within Dom0 (very little memory used by applications). Our OS runs in DomU and is constrained to 128MB of memory. When the system is unresponsive, typing a character into a Dom0 console take 2-5 seconds to appear on the screen. Likewise, other activity is extremely slow. As I mentioned, we have not been able to isolate where the problem is. Running, for example, an OpenWrt Linux build in DomU does not have this problem.
I have seen something similar, though I don't know where the fault lies either.
That is somewhat good to hear. I have today solved this problem by running "xm vcpu-set Domain-0 1." By default, Xen assigned Dom0 all of my cores (two). Reducing this to one solves the problem for me. I am working on a better write up that I'll send to fedora-xen and possibly the upstream Xen mailing list. I have not decided if this is a bug and am having some discussions locally that may help me formulate a better inquiry.
Usually it's better to use dom0_max_vcpus=1 on the grub xen.gz line.
So, is this a known "issue." Is it typically best practice to limit Dom0 to one core? I've seen systems where this is not a problem (dom0_max_vcpus=n works fine, where n is the number of cores) and others where it is. Why would this be?
On Mon, Jan 03, 2011 at 12:05:33AM -0600, W. Michael Petullo wrote:
This kernel works as expected with one exception. The exception has been a nagging problem, but I have not reported it because 1) we are using a research OS in DomU and 2) we are not clear if the problem is in our code, Linux or Xen. But, here are the symptoms:
Occasionally (this seems to correlate to network activity between Dom0 and DomU), the system becomes unresponsive. I am running the Michael Young kernel at runlevel 3 within Dom0 (very little memory used by applications). Our OS runs in DomU and is constrained to 128MB of memory. When the system is unresponsive, typing a character into a Dom0 console take 2-5 seconds to appear on the screen. Likewise, other activity is extremely slow. As I mentioned, we have not been able to isolate where the problem is. Running, for example, an OpenWrt Linux build in DomU does not have this problem.
I have seen something similar, though I don't know where the fault lies either.
That is somewhat good to hear. I have today solved this problem by running "xm vcpu-set Domain-0 1." By default, Xen assigned Dom0 all of my cores (two). Reducing this to one solves the problem for me. I am working on a better write up that I'll send to fedora-xen and possibly the upstream Xen mailing list. I have not decided if this is a bug and am having some discussions locally that may help me formulate a better inquiry.
Usually it's better to use dom0_max_vcpus=1 on the grub xen.gz line.
So, is this a known "issue." Is it typically best practice to limit Dom0 to one core? I've seen systems where this is not a problem (dom0_max_vcpus=n works fine, where n is the number of cores) and others where it is. Why would this be?
Some people want to dedicate cores for vms, so then it makes sense to also limit and pin dom0 vcpus to specific cores.
It all depends on the workload.
See: http://wiki.xensource.com/xenwiki/XenBestPractices
-- Pasi