A litmus check request

[Charles Swiger]

On Sep 8, 2014, at 3:54 PM, Matthew Seaman <matthew at FreeBSD.org> wrote:
> On 08/09/2014 22:37, Dave Babb wrote:
>> I am asking that someone please review this attached fstab and validate
>> for me, that at no time will any compilation intermediate files, or ".o"
>> files will be written to the SSD. I am trying to protect the SSD.
> 
> That should work pretty well.  However what's your swapinfo say?  If you
> do happen to fill up RAM with your tmpfs, you'll end up swapping a bunch
> of memory pages out to the swap area, and if your swap is on the SSD,
> then you've pretty much defeated the object of the whole exercise.

While it's certainly true that being able to keep tmpfs entirely in RAM
is highly desirable, it's likely that hitting an SSD for swap would
generate less wear on the drive than writing compiler temp files directly.

Sure, you don't really want to swap onto an SSD if you can help it, but 
paging and swapping activities are generally page-aligned, and so they
cause less write amplification than filesystem writes are likely to cause.

> That aside, it looks good.  If it's possible, adding noatime to the
> tmpfs mount flags will help with performance.


Sure.  Also feeding -pipe to the compilers will try to use pipelines
between the stages of the toolchain rather than temp files.

Regards,
-- 
-Chuck