Supercomputing 2014 – A Personal Tour d’Horizon of the Most Important Trends

Supercomputing 2014 (SC’14), New Orleans, LA: A (subjective) tour d’horizon of the most important trends / developments from this year’s SC conference.

  • «Data Science vs Compute Science»: On the one hand, high-performance computing (HPC) systems need to become more «data centric» instead of «compute centric» to address the future challenges in scientific computing (FLOPS as a measure of performance is increasingly put into question). On the other hand, current big data architectures are inefficient in terms of performance and energy consumption compared to HPC systems. But how can we better bring HPC and cloud computing in line? The conference has revealed some interesting approaches from the «fusion» of distributed and parallel file systems (e.g. BeeGFS) up to the vision of huge data centers built from SOC’s (systems on a chip).
  • «The End of Moore’s Law»: The discussions around the foreseeable end of the doubling of the transistor density in computer chips taking place approximately every two years (and thus, the corresponding increase in computer power) have clearly intensified. The end could come already in 2018… However, with this fact people also associate new chances. This becomes apparent when an IBM researcher says: «I am happy if Moore’s law ends soon since then people will finally start to think again about computer architecture which they didn’t since the advent of John von Neumann’s architecture in the middle of the last century…» By the way, a highlight of SC is that people dare to discuss together about the next challenges and problems in scientific computing in an open atmosphere and independent of their origin (at least outside the «exhibitor floor»).
  • «MPI + X»: Message Passing Interface (MPI) is still the de facto standard of inter-process communication among the nodes of HPC systems. However, as a result of novel hardware capabilities (e.g. remote memory access), other parallel programming models (the «X») become increasingly important. Partitioned Global Address Space (PGAS) languages (such as e.g. OpenSHMEM) with very straightforward (!) API’s for one-sided-communication operations are among the most interesting ones.
  • «From Big Data to the Long Tail»: The management, analysis and publication of large, (computer generated), scientific data sets causes headaches everywhere in the world. Funding agencies (such as the NSF or the Swiss National Science Foundation) ask more and more frequently for so called «data management plans»; yet, they are often not enforced, also because no funding is available to provide the required infrastructure. The responsibilities are shifted back and forth… The lack of incentive for scientists to publish research data is generally considered as one of the fundamental underlying problems (the only thing that really counts are scientific journal articles). The interest in this «birds of a feather» event was therefore correspondingly modest… However, the importance of the issue has been recognized at the highest ranks. The plenary invited talk «Meeting the Computational Challenges Associated with Human Health» from the associate director of data science, NIH, made this clear. (Fortunately), this event had attracted a lot more attendees.

Many of the SC’14 topics are of great importance to us (i.e., the ITS (ID) and, in particular, ITS Scientific IT Services) and our costumers (i.e., the researchers) and thus, will continue to keep us busy and challenged. This is the exciting side of our work.

Finally

Beside all the exciting trends, the conference has also shown a very «reassuring side»: Namely that the challenges and open questions around scientific computing are the same all around the world. To this end, SC provides an excellent platform for knowledge exchange.

Post

 

Dr. Thomas Wüst, Group Head Research Informatics, ITS Scientific IT Services (ID SIS)

erstellt am
in News

1 Kommentar für «Supercomputing 2014 – A Personal Tour d’Horizon of the Most Important Trends»

Hinterlassen Sie eine Antwort

Ihre E-Mail-Adresse wird nicht veröffentlicht. Erforderliche Felder sind mit * markiert

Diese Website verwendet Akismet, um Spam zu reduzieren. Erfahren Sie mehr darüber, wie Ihre Kommentardaten verarbeitet werden .