Presentation | 2011-04-22 Method of Parallel Count Processing for avoiding the increasing time of Coherence on Many-core CPU Satoshi KONDOH, Reiko SAKURADA, Masashi KANEKO, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | Recently, the performance gain of CPU has come to rely on the number of cores, because of the heat problem according to the CPU frequency rise. Generally, to use up the performance in multi core CPU, we should program by the style of high parallelism. According to the Amdahl's law, we have done various optimizations, such as minimization of sequential processing part, in order to improve the parallelism in the past. In many-core CPU that will appear in the future, it is understood not to be able to solve the bottleneck by only these techniques by the recent researches. Especially, these researches pointed out that the overhead of cash coherence for cores' synchronization increase and cause bottleneck even if small processing such as count processing exist. To solve this problem, in the past, technique, which each core by thread counts independently and integrates values at the end, was proposed. As other technique, method using Optimistic Lock was proposed too. However, because we use the counter with some condition sentence generally, we should add the value of each core to the shared variable. In this case, overhead for synchronization cause every processing loop in the thread. Then, in this paper, we propose the method that improves parallelism by reducing the frequency of synchronization. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | Parallel Processing / Synchronization / Cache |
Paper # | NS2011-15 |
Date of Issue |
Conference Information | |
Committee | NS |
---|---|
Conference Date | 2011/4/14(1days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | |
Chair | |
Vice Chair | |
Secretary | |
Assistant |
Paper Information | |
Registration To | Network Systems(NS) |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | Method of Parallel Count Processing for avoiding the increasing time of Coherence on Many-core CPU |
Sub Title (in English) | |
Keyword(1) | Parallel Processing |
Keyword(2) | Synchronization |
Keyword(3) | Cache |
1st Author's Name | Satoshi KONDOH |
1st Author's Affiliation | NTT Network Service Systems Laboratories, NTT Corporation() |
2nd Author's Name | Reiko SAKURADA |
2nd Author's Affiliation | NTT Network Service Systems Laboratories, NTT Corporation |
3rd Author's Name | Masashi KANEKO |
3rd Author's Affiliation | NTT Network Service Systems Laboratories, NTT Corporation |
Date | 2011-04-22 |
Paper # | NS2011-15 |
Volume (vol) | vol.111 |
Number (no) | 8 |
Page | pp.pp.- |
#Pages | 6 |
Date of Issue |