[Beowulf] Pony: not yours.

atchley tds.net atchley at tds.net
Thu May 16 09:33:32 PDT 2013

On Thu, May 16, 2013 at 11:45 AM, Mark Hahn <hahn at mcmaster.ca> wrote:

> I found it unilluminating, actually.  don't we all know about power issues?
>> Know? Yes. To this extent? Maybe not.
> OK, I can see that.  to me, power is similar to a number of other issues,
> which provide extreme limits to scaling.  (power, reliability, net,
> perhaps even storage and software complexity).  from a CS background,
> these all have distressing O()-type complexity, so each is clearly
> a hard limit at some scale, given some level of technology.
> maybe I'm just more pessimistic ;)
> I guess I did think the flops power versus even on-chip communication thing
> was interesting.  but isn't the lesson to optimize for dataflow *and*
> sometimes perform redundant computations?
>  forward and with a 20 Mw power budget, an exascale machine's network
>>> would
>>>> consume all the power leaving nothing for computation.
>>> well, that sounds absurd - were they assuming a full-bisection fat tree
>>> of
>>> older (hotter, lower fanout) generation interconnect?
>> Even dragonfly topologies (high-radix routers with low hop count) and a
>> bandwidth taper of 1/2 given current power consumption levels will exceed
>> 20 Mw.
> anyone have an url on this topic?  fundamentally, I've always reasoned
> that since we exist in 3d, our networks need to be a 3d lattice at scale.
> fighting power (flops, comm) seems like a noble, *engineering* fight,
> but fighting our existence's dimensionality is silly...

We already exist in 3d - racks. We limit rack heights due to weight already
(another item to add to your list of limits above).

I understand that you are arguing for 3d is silicon (or other substrate),
but there will still be a practical limit.

I saw some presentation a year or two ago that showed an exascale system as
a sphere with the switch complex in the center. It looked very 60s-ish,

> yes, I would like to demo your new blackhole-based interconnect! ;)

Shouldn't it be wormhole-based? The blackhole-based one will just destroy
the data, no?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.beowulf.org/pipermail/beowulf/attachments/20130516/7b35f8a5/attachment.html>

More information about the Beowulf mailing list