[Beowulf] recommendations on ARM distro?

[Gerald Henriksen]

On Mon, 16 May 2016 00:27:21 +0800, you wrote:

>Very nice reply Gerald, not meaning to nit, but for certain workloads
>I'd emphasize that accelerators make more sense than Intel.

Yes, and more below.

>Power and ARM have some uphill battles ahead of them, but I'm
>optimistic that in the next 2-4 years we're going to see Intel go
>against increasingly interesting products.

The good news for Power and ARM is that Intel is in a little bit of
disarray, and while their new emphasis on Server and IoT is likely a
good thing it both depends on what they mean by that and how much of a
disruption the layoffs cause amongs the employees that stay.

The big advantage the ARM could have is the ability of someone to come
along and take the open design of ARM and create some unique hardware.
Quite unlikely, but how about a SoC designed for accelerators, with
say Infiniband in the SoC and support for say 5 accelerators on one
motherboard?

But, to get back to the question that started this thread, the biggest
problem that both Power and ARM have is getting the software working
correctly.  It is clearly unacceptable that getting something as basic
as LDAP working is a time consuming, trial and error task.

In my likely flawed view neither ARM nor Power will escape from their
current niche markets until they get "affordable" hardware to the
public.

The number of programming languages currently in use, the number of
frameworks, the number of applications available is simply beyond the
ability at this point of any distribution, Linux or *bsd, to test
everything.  Instead it is at best a if it compiles, it ships, with
only the core portions of the distribution getting any testing, and
then really only on the hardware that anyone has access to.

If Power had a motherboard/CPU combination in the $300 to $500 range,
and ARM in the $200 range, then all the people who work on the open
source software would be able to at least consider getting a second
system to play with, to test their software on, etc. and as a result
fix things or at least file bugs.  But as long as Power systems are
into the thousands of dollars, and ARM (or more specifically one of
their partners) is offering a 2, almost 3 year old CPU that is one to
two generations old and the unrealistic price of $300 (*) that aren't
going to get the public to buy their hardware, and thus they won't get
the testing/developing/etc of the software ecosystem.

It's a shame, because while I couldn't afford even my optimistically
priced examples, I think we do need a competitor to Intel to keep
things moving, but I just don't see that Power really wants to be that
and ARM is simply failing to execute.


* the cheapest price announced for an AArch64 board with proper
hardware (ie. RAM slots, SATA support, UEFI boot system) is $300 and
offers an AMD A1100 SoC that was announced just over 2 years ago. This
board is not yet shipping, and the reviews from January indicate that
the A1100 isn't really competitive (in performance/power
consumption/price) against Intel Atom offerings.