Dirk H. Schulz wrote:
This kernel repeatedly crashed the Debian dom0 kernel; I tested that on 2 different servers (one dual PIII, where the crash happened within an hour, and one Dual Xeon 3.2 GHz, where the crashes took a few hours to happen).At the moment this is simply a coincidence (as the Debian dom0 kernels crash every few weeks anyway), but in the case of the self built domU kernel the dom0 kernel crashes looked different: no error output, just a blank screen.
>
After disabling the self compiled kernel everything went back to normal.First question of course is: Is it technically possible that the dom0 kernel crashes are related to the NMBCLUSTERS parameter in the domU kernel?
Unlikely. nmbclusters is a value arbitrarily defined to tweak the pool_cache(9) routines (similar to Linux slab allocator) used for allocation/deallocation of mbufs.
One thing is evident, domU should not crash dom0, whatever happened in the domU. Xen's role being limited when dealing with heavy I/O, I dare say that something is wrong within dom0.
Next one of course: Is this a known problem? Is there a workaround? I would very much love to test NetBSD as domU for high load use.
Well, you do have a workaround, by decreasing the number of mbuf clusters. Yes, it does not sound reasonable.
Perhaps faulty hardware?Does it happen when you change the nmbclusters dynamically? (sysctl -w kern.mbuf.nmbclusters=...)
-- Jean-Yves Migeon jeanyves.migeon%free.fr@localhost