Group Webinar May 6, 2020: Botao Li
From Werner KRAUTH
Revision as of 23:26, 20 May 2020 Werner (Talk | contribs) ← Previous diff |
Revision as of 22:58, 24 November 2020 Werner (Talk | contribs) Next diff → |
||
Line 8: | Line 8: | ||
We have discovered that, in ECMC, time needs not to be global. The concept of local time is then introduced, which allows us to abandon the central scheduler and minimize the propagation of information. Parallel ECMC algorithm, implements the idea of local time, has been designed and benchmarked on multithreaded machines. This algorithm is proved to be correct. And with a CPU of 40 logical cores, the prototype of multithreaded ECMC is more than an order of magnitude faster than its singlethreaded counterpart. | We have discovered that, in ECMC, time needs not to be global. The concept of local time is then introduced, which allows us to abandon the central scheduler and minimize the propagation of information. Parallel ECMC algorithm, implements the idea of local time, has been designed and benchmarked on multithreaded machines. This algorithm is proved to be correct. And with a CPU of 40 logical cores, the prototype of multithreaded ECMC is more than an order of magnitude faster than its singlethreaded counterpart. | ||
- | '''Reference:''' https://arxiv.org/abs/2004.11040 | + | '''Reference:''' https://arxiv.org/abs/2004.11040 see [[paper in Computer Physics Communications| Li_Todo_Maggs_Krauth_2020]]. |
+ | '''Code''' All code is open-source and available on github (see the paper). | ||
'''Live Recording:''' Not available | '''Live Recording:''' Not available | ||
[[Group_WK#Group_Webinar|Return to Group Webinars]] | [[Group_WK#Group_Webinar|Return to Group Webinars]] |
Revision as of 22:58, 24 November 2020
Group Webinar Wednesday, 7 May 2020, 10h30
Speaker: Botao Li (Laboratoire de Physique, ENS)
Title: ECMC with local time and its parallel implementation
Abstract: We have discovered that, in ECMC, time needs not to be global. The concept of local time is then introduced, which allows us to abandon the central scheduler and minimize the propagation of information. Parallel ECMC algorithm, implements the idea of local time, has been designed and benchmarked on multithreaded machines. This algorithm is proved to be correct. And with a CPU of 40 logical cores, the prototype of multithreaded ECMC is more than an order of magnitude faster than its singlethreaded counterpart.
Reference: https://arxiv.org/abs/2004.11040 see Li_Todo_Maggs_Krauth_2020. Code All code is open-source and available on github (see the paper).
Live Recording: Not available