Re: Utility of PV on non-obsolete hardware

[tg%gmplib.org@localhost (Torbjörn Granlund)]

"Greg A. Woods" <woods%planix.ca@localhost> writes:

  Interesting, I've never seen such freezes on any of my systems, though I
  can't say I've hammered them too hard with concurrent jobs (other than
  with "make -j 8" or similar, that is).  For example I have a mildly busy
  mail and SSH server in a domU (with one VCPU) that's has an uptime of
  393 days now.  Its dom0 has the same uptime too.

I've never had such DomU NetBSD problems with any release prior to 8.0.
(I have extremely extensively used all releases from 5.0 to 8.0 in the
same setting.  We're talking millions of similar builds.)

  Whenever I hear of a system going unresponsive these days I think of
  paging problems, and perhaps lack of swap space, but if your same
  workloads continue fine in HVM guests, well that's probably less likely.

It could be VM related.  My DomUs have 512 MiB RAM (14 systems) and just
256 MiB (3 systems).  No swap in either case.  But since builds usually
work even on the 256 MiB DomUs, and fail randomly even on 512 MiB
systems, it seems unlikely.  The builds are essentially the same.  (Test
runs are seeded and vary quite much in resource use, but that's not
where I've seen these crashes.)  These DomU do nothing except build and
whack GMP.

Let's see how the 8.0 HVM DomUs work in the weeks to come.  I made the
switch just last week.  They've run flawlessly thus far.

-- 
Torbjörn
Please encrypt, key id 0xC8601622