On Fri, Oct 1, 2010 at 9:37 PM, William Stein <wst...@gmail.com> wrote:
> On Fri, Oct 1, 2010 at 12:19 PM, William Stein <wst...@gmail.com> wrote:
>> Hi,
>>
>> I've done the final switch as outlined below. Now
>>
>> /disk/scratch -- 3.5 TB common scratch on all machines
>> /home -- 3.5TB common /home on all machines
>> /disk/lmfdb -- 3.5TB disk for the L-functions and modular forms
>> project, on all machines
>>
>> Some other machines have some local faster /scratch.
>>
>> You might find that /home is now more usable, possibly. I will look
>> into replacing the network by fast
>> direct cables, too, now that everything is simple up-to-date Linux.
>
> With the current NFS parameters, /home is very slow. I can change the
> parameters somewhat and get dramatically better performance, but have
> saw some issues with certain parameters.
It turns out that /home is so slow with the sync parameter that it
would take like 12 hours to build Sage!
However, with async (which is the default on Linux), things are so
fast that the old trac bug #8677 that
Tom Coates reported gets triggered:
http://trac.sagemath.org/sage_trac/ticket/8677
I've posted a patch for review there fixing the problem. I've marked
it a blocker, since I'm going to
change the nfs mounts to all use async now (which is massively
faster), and anybody who tries to
build Sage using parallel build will likely run into problems unless
they apply the patch at 8677.
It's just a tiny patch that wraps creating a directory in a try/except block.
-- William
--
To post to this group, send an email to sage-devel@googlegroups.com
To unsubscribe from this group, send an email to
sage-devel+unsubscr...@googlegroups.com
For more options, visit this group at http://groups.google.com/group/sage-devel
URL: http://www.sagemath.org
