is TMPFS still highly experimental?

[Chris Rees]

On 2 October 2011 11:23, Olivier Smedts <olivier at gid0.org> wrote:
> 2011/10/2 Xin LI <delphij at gmail.com>:
>> Hi,
>>
>> On Sat, Oct 1, 2011 at 8:48 AM, Chris Rees <crees at freebsd.org> wrote:
>>> I've also not heard of anyone using it with zfs successfully- it tends to
>>> shrink rapidly.
>>
>> I'm quite surprised with this assertion.  I use tmpfs on my own system
>> and I never see such problem as long as one have sufficient swap
>> space.
>
> The problem here is "sufficient swap space".
>
> I've got 8G of RAM, and 2G of swap (just in case). When the ZFS ARC
> reaches 4G, there's no room for a single byte in tmpfs, even with 2G
> swap free and at least 2-3G RAM free. The swap size must be at least
> the RAM size if you plan on using ZFS and tmpfs. That's a problem for
> me because I'm short on disk space, and there's no point in having an
> enormous swap size (hey, minidumps !) when you already have lots of
> RAM, which is the case in most ZFS installs.
>

It's common practice to set the swap size equal to or more than the
physical RAM size, as well as assumed in many places-- how could you
have a crash dump otherwise?

Unfortunately you probably won't get a 'fix' to allow swap<RAM. It's
not just for 'virtual memory' usage, after all.

Chris