Re: efficient access to "rotational"; new fcntl?

[Jim Meyering <jim@meyering.net>]

Arjan van de Ven wrote:
> On Sat, 19 Sep 2009 10:01:51 +0200
> Jim Meyering <jim@meyering.net> wrote:
>> Yeah, I mentioned I should do exactly that on IRC yesterday.
>> I've just run some tests, and see that at least with one SSD (OCZ
>> Summit 120GB), the 0.5s cost of sorting pays off handsomely with a
>> 12-x speed-up, saving 5.5 minutes, when removing a
>> 1-million-empty-file directory.
>>
>
> likely because you actually reduce the amount of IO; inodes share
> disk blocks; repeated unlinks in random order likely write the same
> block multiple times....

That makes sense.
Maybe cache effects, too?

> btw have you given thought about using threads as part of rm -r[f] ?
> (would make the unlinks of unrelated directories/files asynchronous)

While it is certainly a nicely parallelizable process,
rm usually runs so quickly that I doubt it'd be worthwhile.
If you know in advance that parallelizing a particular recursive
removal would give a significant benefit, it's probably best to do it
via e.g., xargs --max-procs=N.

However, sort *would* benefit, and some UCLA students implemented that
for a term project.  Unfortunately, the project is stalled because the
implementation was not efficient enough, and no one has found the
time to improve it since.