The programme of PPAM 2011 included a workshop entitled “Memory and Data Parallelism on Multi- and Manycore Platforms”. Its organisers aimed to highlight the challenges of making effective use of the increasing computational power of multicore processors, many-core platforms and various types of accelerators.
What topics did the workshop cover?
Adding processing units did not necessarily make an application faster. The workshop examined the widening gap between the computing capacity of parallel processors and the rate at which memory could supply data. Memory-access latency and limited bandwidth could leave otherwise powerful hardware underused.
The way data was stored and retrieved was therefore just as important as the number of processor cores. The workshop addressed the different levels of cache memory, including caches dedicated to individual cores and those shared between cores. It also considered non-uniform memory access (NUMA) systems, where the location of data affects access times.
The workshop also considered the hardware available in 2011. Multicore and many-core processors enabled parallel execution; SIMD units and vector registers supported operations on multiple data elements. GPUs used for general-purpose computing (GPGPU) and other specialised accelerators provided further options, including offloading suitable tasks from the CPU. The emphasis was on using processing resources effectively while accounting for memory constraints.
What kinds of research papers were invited?
The call for papers brought together studies of memory systems, parallel programming techniques and demanding scientific applications. Submissions could address four broad areas:
- Memory behaviour and performance – studying cache use, simulating cache activity and employing tools to evaluate computing performance.
- Algorithms for parallel processors – developing hardware-aware methods, investigating scalability and using SIMD vectorisation on multicore and many-core platforms.
- Accelerators and hybrid computing – working with GPUs and other accelerators, combining different computing resources and examining how practical accelerator-based systems are to use.
- Engineering and scientific simulations – handling workloads that demand substantial processing power and memory capacity in electrical, mechanical, civil and biomedical engineering.
The workshop programme included research on multi-GPU scientific computing, parallel numerical algorithms for multicore processors, and the automatic performance tuning of solvers on shared-memory architectures.
Organisational details
PPAM 2011 took place in Toruń, Poland, from 11 to 14 September 2011. It was the ninth edition of this international conference.
The workshop session was held on 12 September 2011, from 10:50 to 12:30, and was chaired by Michael Bader and Josef Weidendorfer.
The “Memory and Data Parallelism on Multi- and Manycore Platforms” workshop was organised by Michael Bader of the University of Stuttgart, together with Carsten Trinitis and Josef Weidendorfer of the Technical University of Munich.
The original paper submission deadline of 30 April 2011 was extended to 15 May 2011. The organisers provided guidelines on submission format and evaluation criteria. Acceptance decisions were scheduled for 15 June 2011, while abstracts of accepted papers were to be posted on the workshop website. Authors had until 31 October 2011 to prepare the final, camera-ready versions of their papers.
The official programme listed four presentations in the workshop session. The PPAM 2011 post-conference proceedings, published by Springer in 2012 as volume 7203 of the Lecture Notes in Computer Science (LNCS) series, included five research papers under the workshop heading.