William Ahern <[email protected]> writes: > On Mon, Oct 15, 2012 at 11:44:30AM -0700, Wesley PA4WDH wrote: >> >> The host will be Gentoo, i haven't decided about the guest yet. It seems a >> bit overkill to use gentoo there too. The workload would be email with an >> MTA, Secure IMAP, Webmail and maybe some spam filtering. The number of >> emails would be somewhere in the 10s per day or so, so i guess that won't >> be a problem. My main question would be if it's responsive enough to be >> usable. Does anyone have any experience with virtualisation on an atom ? > > Excluding virtualization, it should be more than enough. I'm preparing a > small mail and web server for hosting which will take on much more traffic > than that. The only bottleneck I ran into was mutt loading my 200MB+ mbox > files, but compared to an opteron it's only a few seconds longer. (It's > actually mutt's pokey parsing; using my own custom mbox and MIME parser, I > can load and parse the file as fast as the disk can send it--less than 2s.) > > FWIW, I'm using 2 20GB Intel SLC SSDs, one for /home and the other for /var. > > Throwing virtualization into the mix, though, gives me pause. Even with > hardware vx instructions, virtual hosts still have a huge stack of software > to go through for simple things like disk I/O, and these paths execute a > lot. Like with mutt, there's not much overhead with the Atom to make up for > suboptimal code paths. If you're worried about security, I'd use OpenBSD or > Ubuntu (the former because unmaintained and unpatched it's still a hard > target to break, and the latter because Debian Apt takes most of the burden > out of maintenance.)
With xen and a pv guest, the virtualization overhead is low. I have measured about a 10% loss (not on a soekris) from raw disk in dom0 to file in dom0, and also from file in dom0 to raw 'disk' in domU. So basically I concluded that disk IO suffers about 10% (with no activity in dom0 or other domUs). I have not tried xen on a 6501.
pgpxZwYz5K9zi.pgp
Description: PGP signature
_______________________________________________ Soekris-tech mailing list [email protected] http://lists.soekris.com/mailman/listinfo/soekris-tech
