[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: Transputer - schools
If anyone is interested in cheap low power multiprocessor systems, then you should check out the new servers based on the Calxeda processors. These are currently only 32bit quad core processors, but in the project I am just wrapping up we have demonstrated a 3-10x improvement in performance per watt over Intel based servers, and have ported a stack of standard software to them, including some HPC type apps. One of the partners found some unexpected benefits from the high on chip and interchip communications when implementing a distributed file systems.
Some of the R&D has shown (for cloud computing) that "pods" of 16-32 processors with 3D stacked memory (although silicon interposer is better thermally without too much loss of performance) is about the optimum, and that it is practical to build a 64 core chip as 4 replicated pods with each pod having its own memory interface and a couple of extra memory interfaces for access to large memory (we have also found that a 1G DRAM stacked and used as a very large cache has advantages).
I will be publishing all the final public stuff over the next couple of months.
I have also reported on meetings on Exascale computing. Interestingly, some of the experts commented that once you get over 10,000 cores, not many algorithms scale, and it might be worth going back to some of the work done in the late 1980s/early 1990s to see if there is anything that scales better. To get to Exascale, it is more about scalability than efficiency on each node.
Thus it really is timely to be getting kids - the next generation of programmers - to think parallel to start with. Personally, I still think there is a need for a simple parallel programming language based on the CSP/Occam model. I grew up as an engineer, rather than a programmer, and for me it mimics the design of circuits - interconnected black boxes, with race conditions and glitches being the equivalents (roughly) of deadlocks and livelocks. I also find CSP an easy concept because the behaviour of things in the real world can easily be mapped onto it.
Maybe what we need is "occarm" for the Raspberry PI?
Aspen Enterprises Limited email tony@xxxxxxxxxxxx
tel +44-1278-761000 FAX +44-1278-760006 GSM +44-7768-598570 URL: www.aspen.uk.com
Registered in England and Wales no. 3055963 Reg.Office Aspen House, Burton Row, Brent Knoll, Somerset TA9 4BW. UK
From: Sympa,pkg125 [mailto:sympa@xxxxxxxxxx] On Behalf Of J.B.W.Webber
Sent: 25 June 2013 21:18
To: Richard Dobson; occam-com@xxxxxxxxxx
Subject: RE: Transputer - schools
Thanks for the link to http://computingatschool.org.uk/ , I have just registered.
I have a few possible fingers in this pie - I am in communication with a guy in Germany who time-slices an Arm processor so that one gets up to 16 virtual processors on a credit-card sized board; I am also waiting for a student in the USA to complete his thesis and have the time to talk to me about the Beowulf Raspberry Pi cluster he has produced. I see both of these as being highly relevant to the school multi-processor scenario. I am porting an array manipulation language to them.
My particular interest is in the intersection of agile/concise array programming with parallel processing. I believe that this to be the direction in which I would like to see new developments taking place.
From: Sympa,pkg125 [mailto:sympa@xxxxxxxxxx] On Behalf Of Richard Dobson
Sent: 20 June 2013 18:44
Subject: Re: Transputer
On 20/06/2013 16:50, Barry Cook (4Links) wrote:
> Hi All,
> In case any of you didn't see this ...
> Take a special look at the last paragraph on the second page (link at
> the bottom of the above page).
"If I can get parallel computing into the schools that will be a great achievement, because then we wouldn't get all these kids thinking the world is sequential."
I agree; and a highly concurrent Raspberry Pi-like board could garner a lot of interest - even if to begin with only in the form of a simulator.
May I suggest that those interested in this aspect of educational outreach consider joining the CAS-Online forum**? People working at the architecture end of things are somewhat under-represented there, though there is plenty of hand-waving about "the industry". Discussions about concurrency (beyond the entry-level way one can explore it in MIT's
"Scratch") are conspicuous by their absence. To establish concurrency as a serious topic it needs to be brought to the attention of the examination boards who design the curriculum, as well as to teachers.
CAS has already responded to the new National Curriculum proposals, but needless to say things are still in a somewhat "melting-pot" stage.
** via http://computingatschool.org.uk/