On Monday 07 April 2014 10:41:26 Yasha Karant wrote:
> However, I have not used VirtualBox in a mission-critical server
> environment. If anyone else has experience with the Oracle pre-packaged
> VMs under VirtualBox in a distributed server environment, I would
> appreciate additional observations. On a reasonably well provisioned
> workstation, VirtualBox running MS Win 7 (and previous MS Wins) displays
> no real end user experience degradation (long load or execute times,
> crashes, etc.). However, a mission critical server can experience a
> very different workload than a workstation, and it is important to be
> able to predict significant end user service degradation due to the
> virtualization overhead -- particularly in real-world use (not
> simulations or highly controlled environments).
I doubt you would have serious problems, but it may be important to consider
what the future of a VirtualBox setup might look like compared to, say, a KVM
one. KVM is the most commonly applied solution in Linux clouds and server
spaces, regardless the details of the guests. I've never seen a serious server
farm running VirtualBox and I can't recall ever seeing anyone target
infrastructure components toward it within OpenStack. KVM is still an
adolescent and seems to be headed toward enterprise maturity as fast as people
can code/shell out investment money, VirtualBox appears to have achieved its
"stable" goal.
Based on that, I would say VirtualBox is probably inappropriate, though
perhaps serviceable in the short-term. Note that I haven't addressed basic
features like live migration or scriptable interfaces -- I'm not sure where
VirtualBox is in that regard -- but assuming its close to KVM in features, I
would recommend KVM for farms, VirtualBox for the one-off alternative desktop
experience.
CAVEAT: If its even a single must-have datacenter feature shy of KVM, though,
then its definitely a bad idea -- one of the critical differences between an
enterprise load and a workstation load is that enterprise loads have a
tendency to suddenly scale whenever the thing they are serving gets popular
(and that could just mean popular on campus or within the company, etc.). Not
being able to integrate smoothly into a managed infrastructure would be just
as bad, imo. After dealing with OpenStack (or the VMWare cloud management
stuff, if you like writing huge checks) a bit you'll never want to go back to
babysitting servers. Automated management can be more important than system
stability (though you really don't want to have to make a decision between the
two!).
|