[SIGCIS-Members] High Performance Computing in China

Pierre Mounier mounier at msh-paris.fr
Sun Jan 20 15:09:05 PST 2013


Jon,

Your project is highly interesting. To my knowledge, there is not much  
published on continental Europe – while a few articles appeared long  
ago about the Ferranti Atlas computer in Britain. Regarding Germany,  
you may look at the papers published in the IEEE Annals on Telefunken  
machines. The conference proceedings at the Heinz Nixdorf Museumsforum  
(Hashagen,  Ulf & Raùl Rojas (2000): The First Computers. History and  
Architectures, Cambridge, MA), mentioned by  Mariann Unterluggauer,  
contain a paper on the Ferranti Atlas, but nothing on German  
developments post-1962.

Regarding France, no specific history has ever been written. All I can  
do is to translate a slice from my present manuscript on the History  
of the Plan Calcul :

After the commercial failure of the Bull Gamma 60 in the early 1960s  
(17 units manufactured), French efforts regarding supercomputer  
developments seemed to reach a glass ceiling. The small SEA company  
delivered two units of its Dorothee dataflow machine to military labs,  
but received no further funding to expand on this basis. In the  
context of the Plan Calcul, from 1965 to the early 1970s,  
supercomputer development was the matter of many talks, while IBM,  
Univac and Control Data reigned on HPC facilities. The driving idea of  
these talks was to develop a “European” supercomputer line, in a  
joint effort combining specialists of Bull, then CII with British and  
German manufacturers. The combination of partners varied over the  
years, but the process was always the same : Experts representing  
these companies met, often in Brussels, presented technical reports,  
considered marketing data, and regularly came to the conclusion that  
they would only move forward if governments funded the whole R&D and  
manufacturing expenses. In other words, the massive presence of IBM  
and Control Data determined the high level of the “ticket” to pay for  
entering this niche market. And governments in the 1970s preferred to  
direct their funds to the development of computer networks.

The national champion of the French Plan Calcul, CII (1966-1975),  
focused on making mid-range commercial machines, and carefully avoided  
dispersing its forces in supercomputer projects – despite the  
political rethoric which came with it as dressing. The nearest CII  
approached high-end calculators was its large Iris80, of which a 4- 
processor version was tested by 1973, and abandoned for technical  
reasons regarding connection complexity. A seemingly wise decision, as  
“only” 34 Iris80 biprocessors were sold, in competition against GE  
600s and IBM large /370s. The same goes with Unidata (CII-Siemens- 
Philips), of which even the high-end machines developed by CII cannot  
be considered supercomputers.

Supercomputer developments were resumed in the 1980s, in the context  
of both the intensification of the cold war (Soviet invasion of  
Afghanistan, Euromissile tension and American SDI) and the policy of  
the French socialist administration to foster the electronics  
industry. Various projects experimenting with parallel architectures  
(ISIS, Marisis, Acri, etc.) mobilized teams within Groupe Bull, INRIA  
and CNET (respectively the research institutes for computing and  
telecommunications) and in different companies. One of the major  
problems was the cost of making specific components. After the Soviet  
threat disappeared in the early 1990s, military fundings were cut and  
these projects killed.

Breaking the glass ceiling became possible when Bull – an old company  
seemingly in its terminal stage by 2000, dropped its traditional  
strategy as a “universal computer manufacturer” following IBM, and  
concentrated on designing highly parallel computer systems in  
partnership with the Commissariat a l’Energie Atomique (CEA), buying  
components off-the shelf.

You’ll find the rest of the story on press releases, such as :
http://www.wcm.bull.com/internet/pr/rend.jsp?DocId=660679&lang=en
http://www.wcm.bull.com/internet/pr/rend.jsp?DocId=713261&lang=en

Best wishes.
Pierre

Pierre Mounier-Kuhn
CNRS & Université Paris-Sorbonne
http://pups.paris-sorbonne.fr/pages/aff_livre.php?Id=838




Le 19 janv. 13 à 00:01, Jon Lindsay a écrit :

> Dear Colleagues,
>
> I'm part of a project at UCSD, now in its very early stages and  
> funded through Lawrence Livermore NL, to assess Chinese progress and  
> prospects in High Performance Computing. As you probably know, China  
> has recently exploded onto the HPC scene with 13% of the Top500  
> systems and growing, including a brief showing at the top with the  
> Tianhe-1A. An article in the most recent Science even suggests that  
> China may be the first to hit the exascale mark. (Incidentally, if  
> this happens and if it happens on Moore's timeline, it will raise  
> some very interesting questions about just how the heck Moore's Law  
> is working across international boundaries and quite different  
> cultural approaches to S&T).
>
> We are very much interested in the institutional context & drivers  
> of Chinese HPC growth, and in comparing this with the history of  
> supercomputing elsewhere (US, Europe, Japan). I would be very  
> interested in any work you might recommend on this history. On the  
> US experience I am aware of MacKenzie's excellent 1991 and 1994  
> articles, books on Pentagon support by Norberg & ONeill 1996 and  
> Akera 2008, NRC volumes on HPC from 1999 and 2004, and Edwards 2010  
> is certainly relevant too. I also see that there's a volume on the  
> history of the Top500 by Meuer et al coming out later this summer  
> which appears to emphasize mainly technical milestones. What other  
> key texts on the evolution of supercomputing am I missing,  
> especially beyond the American case?
>
> Thanks very much for any and all suggestions!
>
> Best,
> Jon
> _______________________________________________
> This email is relayed from members at sigcis.org, the email discussion  
> list of SHOT SIGCIS. The list archives are at http://sigcis.org/pipermail/members/ 
>  and you can change your subscription options at http://sigcis.org/mailman/listinfo/members

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.sigcis.org/pipermail/members-sigcis.org/attachments/20130121/a37afc89/attachment-0001.htm>


More information about the Members mailing list