On 29 August 2014 15:36, Ewan MacMahon <[log in to unmask]> wrote:
>> -----Original Message-----
>> From: Testbed Support for GridPP member institutes [mailto:TB-
>> [log in to unmask]] On Behalf Of Jeremy Coles
>>
>> In that case if anyone has suggestions/input for the pre-GDB then perhaps
>> they can email it please. I'm not sure if the meeting itself will have
>> Vidyo (but it should).
>>
> Well, looking at the agenda, the one bit that particularly grabbed my attention
> was:
> "Target date and milestones towards cloud as CE replacement"
>
> Last I'd heard, we hadn't particular decided to replace our CEs with cloud
> interfaces, much less be in need of a target date and milestones. Moreover,
> while I think a vac style system could potentially be a replacement for the
> traditional 'push model' infrastructure, a full cloud system (e.g. OpenStack),
> can't - they have the same problem as the traditional approach of being push
> based, they add on some specific problems of their own, and they're a lot more
> complex to set up.
>
This, almost entirely (with the additional note that there are also
several other ways to start up VMs, if that's really what the
experiments want).
But certainly:
I am surprised that anyone is seriously talking about "cloud as CE
replacement" at this point in time, especially given the distinct
differences between "proper Cloud" infrastructures and interfaces and
CEs (and batch systems - there's a particular problem here that
there's a fuzzy line between CE+batch system and "Cloud", which is
being unhelpfully swept over here).
As Ewan notes, Cloud interfaces are not a magical panacea for
perceived or actual problems with CEs, and Actual Cloud Systems
actually have quite a lot of the same properties as WMS+CE style
interfaces (for the same perfectly good reasons), which I had thought
the Experiments invented Pilots to get around. (VAC at least has the
advantage of not actually being a Cloud system, and therefore being
closer to what the Experiments appear to desire in a VM management
system.)
> I think it makes a lot of sense to have enough cloud experience within WLCG
> that we can take advantage of other people's clouds when we can, but it makes
> no sense to run a full cloud stack on WLCG dedicated hardware; it's a lot like
> what we do now, but worse.
Agreed.
Sam
>
> Ewan
|