Intel RSD Limits

[Flint WALRUS]

Hi Andres, thanks for taking times to answer my questions.

Regarding the limitations, I was looking for intel documentation indeed,
not the maas one as I assumed that maas didn’t hardcoded anyone.
So, if you’ve got some links to share I’m all in :-)

About my question regarding the ressources aggregates.
I didn’t get it. How is this supposed to work? Does RSD systems are
provided with a specific OS/hypervisor dedicated to create VM that can be
hosted on two different hosts but still be seen as a unique ressources
within the guest OS? Is that using MPI under the hood?
How can I create a node that will address ressources across multiple
physical machines?

Thank a lot again!
Le lun. 16 oct. 2017 à 23:25, Andres Rodriguez <
andres.rodriguez at canonical.com> a écrit :

> Hi Gael,
>
> The limitations of MAAS wrt Intel RSD support are given by RSD itself. As
> of today, it is not yet possible to slice amount of CPU/RAM/Local storage a
> machine inside Intel RSD, as this is basically on actual systems that are
> part of Intel RSD itself.
>
> Further comments inline:
>
> On Tue, Oct 10, 2017 at 3:36 AM, Flint WALRUS <gael.therond at gmail.com>
> wrote:
>
>> Hi guys, looking at the pods and so Intel RSD limitations but I hardly
>> found documentation about it.
>>
>> So, what currently are the limitations? Is an aggregated machine limited
>> by the amount of hardware that it can request?
>>
>> Imagine that I’ve got 4 RSD Servers equipped like:
>>
>> 4x socket Intel 16 Core CPU.
>> 512 GB Ram.
>> 56 TB SSD.
>>
>> Does it means that I can aggregate the 64 cores with 2TB RAM and 224 TB
>> storage?
>>
>
> Yes, that is correct, however, due to RSD limitations, once you request a
> machine with 10 cores, 200GB RAM & 20 TB of storage disk, RSD will give you
> one of the physical systems that at least match your request. In other
> words, it will give you 1 of the 4 servers.
>
> Hope this answers your question!
>
>>
>> If that’s correct how is that possible? Using a hypervisor with MPI
>> capabilities  ?!
>>
>> If that’s incorrect what are the limits? Am I limited to the amount of
>> resources that can provide a standalone node? If so, why would I use pods
>> instead of classic BM with juju on top of maas being able to use lxd
>> isolation for resources optimization ?
>>
>> Thanks a lot guys!
>>
>> --
>> Maas-devel mailing list
>> Maas-devel at lists.ubuntu.com
>> Modify settings or unsubscribe at:
>> https://lists.ubuntu.com/mailman/listinfo/maas-devel
>>
>>
>
>
> --
> Andres Rodriguez
> Engineering Manager, MAAS
> Canonical USA, Inc.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.ubuntu.com/archives/maas-devel/attachments/20171016/cf57941e/attachment.html>