Re: [Publicity] Clear Containers - Intel's KVM/KVMTool based container distro - what are the implications?

[Anil Madhavapeddy <anil@xxxxxxxxxx>]

> On 23 May 2015, at 09:56, Richard Mortier <richard.mortier@xxxxxxxxxxxx> 
> wrote:
> 
> On 22 May 2015 at 17:19, Lars Kurth <lars.kurth.xen@xxxxxxxxx> wrote:
>> To provide a preview of the results: we can launch such a secured container
>> that uses virtualization technology in under 150 milliseconds, and the
>> per-container memory overhead is roughly 18 to 20MB (this means you can run
>> over 3500 of these on a server with 128GB of RAM).
>> 
>> ---
>> 
>> I am wondering how the typical unikernel compares in terms of start-up time
>> and memory overhead.
> 
> Define "typical" :)
> 
> For MirageOS, I think those are comparable, though MirageOS can go
> lower in terms of memory use (16MiB used in NSDI'15; on-disk size of a
> presentation unikernel typically under 10MB -- NSDI'15 presentation
> claims 2MB) and faster in terms of startup (20-30ms claimed on x86 in
> the NSDI'15 paper, and 30-45ms claimed in the presentation -- Anil, is
> there really a discrepancy?).

I think the discrepancy was due to SYN caching or just a clean boot.
I wonder if their launch times also include setting up network
bridging as well. One benefit of unikernels is that we dont have to
setup a union mount filesystem to boot -- so no touching disk.

-a


_______________________________________________
Publicity mailing list
Publicity@xxxxxxxxxxxxxxxxxxxx
http://lists.xenproject.org/cgi-bin/mailman/listinfo/publicity