Re: [PATCH 1/4] Virtualization/containers: introduction

[Sam Vilain <sam@vilain.net>]

Kevin Fox wrote:
>>>The container is just an umbrella object that ties every "virtualized" 
>>>subsystem together.
>>I like this description; it matches roughly with the concepts as
>>presented by vserver; there is the process virtualisation (vx_info), and
>>the network virtualisation (nx_info) of Eric's that has been integrated
>>to the vserver 2.1.x development branch.  However the vx_info has become
>>the de facto umbrella object space as well.  These could almost
>>certainly be split out without too much pain or incurring major
>>rethinks.
> How does all of this tie in with CPU Sets? It seems to me, they have
> something not unlike a container already that supports nesting.

Yes, I saw that.  AIUI that's mainly about binding groups of processes
to CPUs, to defeat Amdahl's law when it rears its head.  It fits into
the containers model as a hard partitioning feature, but is a lot more
crude than the CPU Token Bucket scheduler in Linux-VServer.

No doubt both CPU allocation strategies will be useful.

Sam.