Re: linux dom0 versus NetBSD dom0

[Konrad Rzeszutek Wilk <konrad.wilk%oracle.com@localhost>]

On Wed, Jan 11, 2012 at 11:21:23PM +0100, Roger Pau Monné wrote:
> 2012/1/11 Brian Buhrow <buhrow%lothlorien.nfbcal.org@localhost>:
> >        Hello.  I realize the subject of this message could be viewed  as
> > inflamatory, but I hope people won't take it that way.
> >
> >        As part of my day job, we're looking at implementing a large Xen
> > installation.  We are a long time NetBSD shop, but for this project, we're
> > trying to evaluate the pros and cons of selecting Linux versus NetBSD as
> > the platform on which to build the core of our infrastructure.  As part of
> > that process, I have the following questions for this group:
> >
> > 1.  Does anyone have i/o bench marks comparing network and disk throughput
> > when using Linux versus NetBSD as Dom0?
> 
> There's an old disk benchmark that compares Linux vs NetBSD Dom0
> performance, it's really old, so I don't know how much of this is
> still true:
> 
> http://lsi.vc.ehu.es/pablogn/docencia/configuraciones/xendom0/xendom0.html

I can tell you that comparing 2.6.18 vs 2.6.32 vs 3.0, you will see dramatical
performance improvements in 3.0. Lot of things were fixed when SSDs started 
coming
in the market with very low latency numbers.

Honestly I don't know NetBSD 5.1 vs say Ubuntu 12. I think your best
bet is to run the same benchmark (fio?) on the same machine and see how it works
out with both.

For guest backend I would recommend Logical Volumes (LV). It can be on top
of an iSCSI disk or so. In terms of NFS, not sure.

> 
> > 2.  We know we can use the xbd driver
> > to provide virtual disks in NetBSD, but are other storage technologies
> > mature in the NetBSD environment?  I'm thinking aboutthings like ISCSI,
> > NFS, or other network storage technologies.
> 
> Storing DomU images in a NFS is a no go for NetBSD, trying to boot a
> DomU that has it's disk on a NFS will make the NetBSD Dom0 kernel
> crash. Here is the related PR:
> 
> http://gnats.netbsd.org/40726
> 
> With no solution (I've tried this with a September build of -current,
> and the problem is still there, I don't know if this has been fixed in
> the meantime). You can however use iSCSI, as reported by Stephen
> Borrill:
> 
> http://mail-index.netbsd.org/port-xen/2011/10/31/msg007039.html
> 
> Starting and stopping machines can make -current kernel crash (from
> September), will try to submit the relevant PR tomorrow if it also
> happens with present version of -current, because I've completely
> forgot about this.
> 
> Also the only image format available on NetBSD is raw or physical
> partition (LVM included), qcow, qcow2 or vhd are only available on
> Linux right now, due to the fact that NetBSD has no grant table device

Well, the upstream kernel does not have blktap either.

> that allows access to grant references from userspace programs (and
> also has no blktap support).