swap partition > 4G

[Poul-Henning Kamp]

In message <20030714170024.GA27936 at HAL9000.homeunix.com>, David Schultz writes:

>As far as moving the responsibility for striping swap to ccd(4) is
>concerned, that sounds like a step in the right direction.
>However, it's a completely orthogonal issue.

Fully agreed.

>The ~150 bogus lines
>of code that could be killed have nothing to do with the size
>limitation.  Also note that ccd(4) can't fix all of the bogosity.
>For instance, you're still stuck with static striping, and you
>have to pretend that (MAXDEVS - curdevs) packs are full.

First of all, there is little point in striping things the way it
is done today, it would probably be equally efficient, if not more
so (due to larger chunks being possible), to concatenate the swapon'ed
devices and allocate from the in a round-robin or even, now that
we have the statistics supporting such decisions, based on
disk-busyness.

>I think
>it would be even cooler if a more general and complementary
>interface to ccd(4) were developed that provided a generic way of
>allocating and deallocating virtually-addressed disk blocks from a
>pool of storage, but I don't have the code for it.  :-P

It's called FFS, and we don't want it for swap, in particular
we want something which does not need a lot of RAM _or_ disk
reads to figure out which bits are busy and which are not.

-- 
Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20
phk at FreeBSD.ORG         | TCP/IP since RFC 956
FreeBSD committer       | BSD since 4.3-tahoe    
Never attribute to malice what can adequately be explained by incompetence.