perform builds on a different drive

[Dominic Fandrey]

On 11/09/2010 21:00, Dmitry Morozovsky wrote:
> On Sat, 11 Sep 2010, Dominic Fandrey wrote:
> 
> DF> > FWIW, I tried tmpfs today on decent 8-STABLE/amd64, and twice put my tinderbox 
> DF> > machine into
> DF> > 
> DF> > panic: tmpfs_alloc_vp: type 0xffffff000d957c40 0
> DF> > 
> DF> > (didn't dig into this deeply enough to get backtraces yet; to be done later)
> DF> 
> DF> This is really interesting. My /tmp is a tmpfs on every machine
> DF> and I don't have such panics.
> DF> 
> DF> Did you run memtest? I used to have RAM problems and they only showed
> DF> up in memtest, when I ran several instances at once.
> DF> 
> DF> I.e. 4 2gb instances on a machine with 8gb RAM.
> 
> Of course, I'm pretty sure my hardware is OK, since it survive both 12+ hours 
> memtest86+ and 12-way buildworld (this machine is also builder for our usual 
> archs, which are stable/releng-[678] for both i386 and amd64, plus various 
> security branches and some local patched ones).
> 
> /tmp is on usual mdmfs, and I sometimes nullfs-mount subtrees of /b, which is 
> tmpfs, under /usr/local/tb/<buildname>
> 
> For the reference, it's Supermicro with Core2Q-9450/8G/3ware/ZFS.  FreeBSD 
> builds are scheduled (NO_CLEAN if exit code is zero, so usually it completes 
> quickly) twice per day; the only panics I encounter last few months are tmpfs 
> related.
> 
> Did you try stress you memory with both tmpfs and other kernel-related tasks?
> 

No I didn't. I might have to do more extensive testing.

Regards