[Beowulf] Hyper Convergence Infrastructure

[Joe Landman]

On 10/03/2015 06:18 AM, Lechner, David A. wrote:
> Hi
> I am wondering if anyone on this list has benchmarked the impact of an 
> HCI solution on performance, or how this newest "next big thing" 
> compares to a new Linux/intel commodity solution?

[disclosure: we are in this space, so take anything I write with the 
appropriate caution]

As with everything, there is a marketing bit and a real aspect. 
Marketing HCI is another way for people to sell "similar" kit to what 
they've done in the past, packaged in a different manner.

Real HCI is about driving efficiency, density, and performance way up, 
while providing a simple mechanism to manage, run, analyze, and tune.

Software HCI is about glue code that you can, in theory, turn a 
pile-o-rusty-pcs into a high performance and highly reliable system.

There are quite a few Marketing HCI "solutions" out there.  Most of them 
are repacked/warmed over systems that don't do terribly well in real 
world use cases.  (note:  saying this from a competitive point of view, 
so feel free to dispute/argue this) In many cases they package some sort 
of commodity box, with lower end kit with a number of things to make a 
"HCI" stack.

In short, if they are not providing their own high performance data 
fabric, they aren't real HCI.  And no, 10GbE is not a high performance 
data fabric (IMO anyway).

Real HCI is all about driving performance, density, and efficiency way 
up.  This means you can do (far) more with often (far) less kit.   Real 
HCI should let you run bare-metal, containers, or VMs. There are caveats 
with all of these, depending upon use case.

> Is there some performance penalty fron the virtualization?

Additional layers are almost never free.  Containers might be better for 
HPC apps, though you will lose the isolation that VMs give you. If you 
are doing VMs with virtual function pass through, chances are that a 
container would be a better (performance) choice, though not one from a 
safety standpoint.

> How do price points per FloP compare?

... to what?

> Is the advantage in the Systems administration, and is there a 
> comparable open source solution?
>

There isn't a complete HCI stack thats open source in the linux space.  
Some might point to OpenStack and others, but really they are targeting 
more of the cloud side (maximizing tenancy per system, as compared to 
application performance and density that real HCI targets).

Beowulf is/was about targeting inexpensive and self controlled 
performance.  Freeing a computationally intensive worker from the 
confines of a particular set of vendors, and enabling them to innovate 
parallel solutions to their problems in a cost effective and efficient 
manner.  Real HCI is about leveraging very performant, dense and 
efficient systems to solve a number of problems (not typically HPC, but 
no reason why not).  Cloud systems are generally about maximizing 
tenancy/utilization of resources, and not so much about individual 
application performance).



> Thanks in advance for  any insights.
> Dave Lechner
>
>
> _______________________________________________
> 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.
e: landman at scalableinformatics.com
w: http://scalableinformatics.com
t: @scalableinfo
p: +1 734 786 8423 x121
c: +1 734 612 4615