[Beowulf] Gentoo in the HPC environment

[Joe Landman]

On 06/28/2014 09:53 AM, Jonathan Aquilina wrote:
> I am going to take this thread down another possible road that nobody has
> mentioned.
>
> What about an HPC cluster in a data center, enterprise, environment hell
> even an ISP environment. Does the same still apply?

Very much so ... you need the flexibility to choose as many of the 
system features as you can.  The OS and distro have been reduced to a 
feature detail, in the sense that they will add/subtract 
complexity/performance/security.  As with other details, there is always 
a cost-benefit analysis one can work through.

You list the features you require, the features you have a strong 
preference for, features you would like, and examine the costs/choices 
you need to make to achieve these.

If you can easily manage/deploy/control gentoo, without significant 
pain, and it meets all your other requirements, solves your needs 
without causing you additional grief/time/effort/resources, use it.  If 
it causes you pain, examine whether or not that pain is warranted or 
needed.  This is the choice we made with CentOS/Red Hat a while ago, and 
Windows long before that.

Use what makes sense, though the definition of "makes sense" varies per 
context.

We have large customers running an OpenSolaris rebuild as the basis for 
their global cloud.  It makes sense to them (engineering, support, etc.).

Regardless of this our tiburon/Scalable OS system supports booting 
anything over PXE (yeah, DOS included ... not kidding at all) to bare 
metal, VMs, etc. so its not an issue for us.  The OS is a detail, what 
matters is how you use it and how you can use and support it.  It 
doesn't matter where it is, what it is, and whose using it.  This isn't 
an ego issue, its a pragmatic issue, which is why flexibility in all 
things is what matters most.  It gives you the greatest ability to 
optimize the things deemed important, while reducing the costs of these 
choices to manageable levels.  You make a change when the costs of the 
choice exceed the value of the choice.

>> Instead of letting this devolve into a distro battle (I have no dog in
>> that race, but I know from long hard experience what to avoid), it makes
>> more sense to look at the bigger picture.
>>
>> In the larger frame, a cluster is a mechanism to provide computing
>> cycles.  The keepers of the cluster are service folks, in the sense that
>> they are providing a shared resource with specific functionality, and
>> providing a service to the internal (and sometimes external) consumers
>> of the service.
>>
>> In this day and age of software defined everything, a cluster needs to
>> be as flexible as possible, and provide the necessary level and type of
>> service to be viable.  Not simply economically viable, but practical,
>> and pragmatic.
>>
>> Which means cluster admins and service teams need to address many
>> different environmental issues and requests.
>>
>> In academic circles, where there may be less of a push for commercial
>> support on software, these requirements may be relaxed relative to other
>> users.
>>
>> In commercial circles, where one might need to guarantee results (for
>> any number of reasons, and yes, this happens), the environments are far
>> more rigid.
>>
>> How can a provider of cycles provide service to a rigid set of
>> requirements without being flexible?
>>
>> My argument is, fundamentally, that technologies like kvm, and Docker on
>> Linux provide a simple mechanism for that functionality.  On Windows
>> (very few windows clusters, but still) you can do this with HyperV.
>>
>> So the details of what runs at the base level on the cluster matter far
>> less than the detailed requirements and the business needs for the
>> application.  The latter should determine the former, and if the latter
>> requires something different than the former supplies, kvm/Docker etc.
>> can provide this.  So can bare metal stateless.
>>
>> Or conversely, you could simply provide exactly one type of computing,
>> and watch your users go elsewhere, specifically to resources that will
>> give them what they require.  Somehow that seems to be not-precisely
>> what this crowd would want though.
>>
>> Its just a thought though.  Gentoo or not doesn't matter as much as
>> *how* your users need to use it.  Thats the point of pain.  If the
>> distro can't handle it, or isn't supported correctly, yes, you'll need
>> to change.  If your cycle provider is rigid in what they will provide,
>> its pretty easy to go to another cycle provider.
>>
>> This is what clusters in clouds have created.  This is why there are
>> folks like Cycle Computing for cloud based clusters, and many good folks
>> like Sabalcore with bare metal systems.  Application and business needs
>> dictate platform choices.
>>
>> --
>> Joseph Landman, Ph.D
>> Founder and CEO
>> Scalable Informatics, Inc.
>> email: landman at scalableinformatics.com
>> web  : http://scalableinformatics.com
>> twtr : @scalableinfo
>> phone: +1 734 786 8423 x121
>> cell : +1 734 612 4615
>> _______________________________________________
>> Beowulf mailing list, Beowulf at beowulf.org sponsored by Penguin Computing
>> To change your subscription (digest mode or unsubscribe) visit
>> http://www.beowulf.org/mailman/listinfo/beowulf
>>
>
>


-- 
Joseph Landman, Ph.D
Founder and CEO
Scalable Informatics, Inc.
email: landman at scalableinformatics.com
web  : http://scalableinformatics.com
twtr : @scalableinfo
phone: +1 734 786 8423 x121
cell : +1 734 612 4615