[Beowulf] Docker vs KVM paper by IBM

Joe Landman landman at scalableinformatics.com
Tue Jan 27 14:27:45 PST 2015

On 01/27/2015 10:33 AM, Jason Riedy wrote:
> And Andrew Holway writes:
>> The most interesting subject around docker is security and the fact
>> that it provides pretty much null actual "containerisation"
> I know I'm more interested in it for "packageization:" Provide

This is the problem that I think everyone using Docker now is looking to 
solve.  How can you distribute an app in a reasonable manner an remove 
all of the silliness you don't need in the app distribution that the 
base OS can solve.

If anything I expect Docker et al to change more on the distribution 
side of things.  You no longer need to care what level of libs your core 
OS provides, you can safely/effectively ignore that.  You can now 
provide something akin to a preconfigured and "working" 
micro-environment that people can trivially deploy.

This is why Docker is so interesting.  But it changes the dynamics of 
the base distribution to be now an orchestrator/launcher/service 
provider rather than a self contained unit of install.

That means distros will need to rapidly adapt to this change 
(Ubuntu/RedHat have for the most part), though the question of how you 
monetize support for an operating system largely stripped of its 
previous core functions should be interesting to see evolve.  I think 
this will be the orchestration management and storage side that gets 
more interesting.

> and support a very low-level, bare OS, then let different apps
> build an environment on top of it.  That eases partitioning
> support work between the stack of app libraries and lower-level
> interfaces.


> On a cluster, it'll likely be one Docker thingy (or maybe
> Rocker...) running on multiple, whole nodes.  I'm not worried
> about isolation between containers on one machine.  These

Containerization purists worry about that.  I like the Packer-ization 
concept more than anything else for clusters and clouds.  Make things as 
easy and fast to startup as possible.  Take away the install step from a 
deployment.  Make it an on button.

> containers will have direct access to GPUs, IB, etc.  Now there
> may be some nifty things you can do for playing with a virtual
> ethernet at L2 that lets containers have access they otherwise
> wouldn't, but that's more for research...
Joseph Landman, Ph.D
Founder and CEO
Scalable Informatics, Inc.
e: landman at scalableinformatics.com
w: http://scalableinformatics.com
t: @scalableinfo
p: +1 734 786 8423 x121
c: +1 734 612 4615

More information about the Beowulf mailing list