Tmpfs elimination of double-copy

[Bruce Evans]

On Tue, 22 Jun 2010, Alexander Leidinger wrote:

> Quoting Kostik Belousov <kostikbel at gmail.com> (from Tue, 22 Jun 2010 11:10:05 
> +0300):
>
>> On Tue, Jun 22, 2010 at 09:13:40AM +0200, Alexander Leidinger wrote:
>
>>> Did you measure the performance before/after? If not, what are your
>>> performance expectations? I don't expect we get double the
>>> performance, but if every data of a write is copied twice, I would
>>> guess there is a measurable benefit.
>> No, I did not bothered. Real benefit of the change is the memory saving.
>
> For me the real benefit is that it survives a fsx run now. Anyone can buy 
> more money and faster machines, but stability...

It's not so easy to buy machines enough faster to compensate from thrashing
of caches caused by extra memory accesses.

> This does not mean I do not appreciate the memory saving (when the change 
> hits one of my machines, I may decide to use tmpfs in places where I didn't 
> use it before because of memory size concerns).
>
> That being said, I'm sure that mentioning the performance aspect additionally 
> to the fsx and memory parts may be good in the release notes (and/or a 
> blog/whatever post of someone).

How much performance does it give anyway?  I would guess a negative amount
compared with a an async mounted ffs, at least if it double buffers
everything, since the double buffering would halve the amount of memory
available for caching files.

Bruce