Neil Bothwick <neil <at> digimed.co.uk> writes:

> 
> 
> On 7 January 2015 13:16:03 GMT+00:00, Rich Freeman <rich0 <at> gentoo.org>
wrote:
> On Wed, Jan 7, 2015 at 3:36 AM, Neil Bothwick <neil <at> digimed.co.uk> wrote:
>  Personally, I wouldn't bother, there is not that much of a gain when
using tmpfs, so if you want to keep all the working files after compilation,
the extra overhead and complexity of copying to hard disk would make it not
worthwhile. Just stick to a spinning disk.

Hmmmm.


> I've found that compiling on tmpfs has a fairly significant performance
gain, but you'd completely negate it of you copied everything to a hard
drive anyway.When you build on a hard drive the file system is going to
treat all those intermediate object files with great care and ensure that
they aren't lost in the event of a power failure, forcing them to be
committed to disk within 30 seconds (typically) and blocking IO

OK.

> when that happens.  Then a minute later it is going to go and have to
delete all those files it so carefully committed. When you build on tmpfs
nothing gets written to your file system except for the final output of the
build.  All those intermediate files are deleted having never blocked your
disk IO.  However, if your goal is to actually save them anyway, then you
might as well just build on disk - moving files doesn't take all that much
IO on most file systems.If your goal is to create tarballs of the saved
builds then you're probably still better off building on tmpfs and creating
the tarball via a hook or something.

Yea, I could not rationalize a solution either, hence the post.


> That's why I thought XFS may help. 
> Reports of the speed gain from tmpfs are quite mixed, but I do use it myself. 

I'm moving to btrfs and eventually ceph, so xfs is not on my roadmap....

I do not have performance problems (FX8250 with 32G ram). And eventually
I'll stumble into something robust with clustering. (mesos-distcc)[1] and
others efforts will yield what I desire (eventually) or something very fast
and close enough [2].


I had some overall resistance to the idea too, but I thought it best to
query the list to see if something robust and simple did already exist....


Thanks for all the posts,

James


[1] https://github.com/mesos/mesos-distcc

[2] http://www.zentoo.org/   (Continuous Integration)





Reply via email to