Things to do this week instead of arguing about mixers

[Eric Sandeen]

Tom Lane wrote:
> John Reiser <jreiser at BitWagon.com> writes:
>> Eric Sandeen wrote:
>>> Now that we have ext4 as the new default filesystem, it'd be nice if we
>>> can get more applications to take advantage of some of the features.
>>   [snip]
>>> Only a few applications are taking advantage of [preallocation] so far,
>>> in part because it's new.
> 
>> And in part because the compatibility issues are a pain.
> 
> And in part because they don't give a damn.  It's really hard to name
> very many applications where this is likely to give a big enough
> performance boost to justify dealing with filesystem-dependent
> optimizations.  (Because it's filesystem dependent, you couldn't even
> drive it off a configuration switch; you'd have to probe at runtime
> to determine whether there was anything you could do.)

nah, if it was detected at build time, you can just call fallocate.  If
you get back ENOSYS, carry on.

Really, even if you just ignore all errors, you're no worse off than if
you hadn't called it.

> As an example: you don't seriously believe that this is the bottleneck
> for downloading RPM files, do you?

No, of course not, and I'm not suggesting that it be used for rpm
downloading.  You wouldn't want to use it for every little file you
write, either.  But for some cases it can be a big win.

Torrent downloading?  This is sort of the quintessential case where it
can help.  Databases?  yes.  Rsyncing large files?  yep.  Creating
virtual images? yep.  Helping samba cope with weird windows client
behavior?  yep[1]

Basically anything that is filled in over time, or filled in sparsely,
could potentially benefit.

-Eric

> 			regards, tom lane
> 

[1]
http://software.intel.com/en-us/articles/windows-client-cifs-behavior-can-slow-linux-nas-performance/
this one I have a patch for, though :)