Re: [gentoo-amd64] What should the swap size be for 4G ram?

[B Vance]

On Sat, 2007-09-22 at 17:22 +0200, Volker Armin Hemmann wrote:
> On Samstag, 22. September 2007, Richard Freeman wrote:
> > P.V.Anthony wrote:
> > >> If following the old rule, 8G should be allocated for swap. I feel
> > >> that is too much. Does 2.6 kernel really need so much of swap with 4G
> > >> of ram?
> > >>
> > >> Was thinking of just using a 1G swap file for safety. Please share
> > >> some thoughts on the this swap size issue.
> > >
> > > Please ignore this email. It looks like I have asked something similar
> > > to this before. I will read the old thread.
> >
> > That's Ok, I got a chuckle out of it.  You have Duncan who doesn't use
> > swap at all (I think), and you have me:
> >
> >              total       used       free     shared    buffers     cached
> > Mem:       2058448    2041388      17060          0      82084     420860
> > -/+ buffers/cache:    1538444     520004
> > Swap:     17514480    1152972   16361508
> >
> > I guess there is a happy medium.  But what else am I going to do with
> > that odd space that doesn't fit easily into a RAID-5?  I figure that if
> > the kernel can find a use for it I might as well let it...  :)  I
> > probably have 50G more of unpartitioned space lying around since I've
> > installed my 2 RAID-5s on non-identical drives.  I guess I'll just have
> > to wait until ZFS takes off on linux...  :)
>
> why? zfs is slow and is mixing things that should be in different layers. One
> argument against reiser4 always was 'it violates the layering' - well this is
> even more true for zfs.
>
> And from this numbers:
> http://tastic.brillig.org/~jwb/zfs-xfs-ext4.html
>
> it doesn't look so great.
Maybe it's just me, but I tend to take comparisons of apples and oranges
with a salt block.  To make that a useful comparison, the kernel
parameters would have to be shown as well as the level of support for
the various hardware.  The second is his write-up does not match his
numbers.  He writes ZFS "...has very bad sequential transfer with
hardware RAID and appalling sequential transfer with software RAID" but
fails to mention how equally bad EXT4 was at the same task.  XFS does
appear to blow them both out of the water in this area.  The rest of the
tests have ZFS being roughly equal to (or in some cases better then)
then the other two options.  

Maybe the next time, the test will be run using the userland ZFS on the
same system as the EXT4 and XFS system.  At least that way you can
adjust the numbers for the overhead of userland talking to the kernel.
Apples to Apples with a handicap is far more accurate then apples to
oranges.

I don't know enough about the layers you mention so I can't say whether
that's a real problem or not.



-- 
gentoo-amd64@g.o mailing list