Analyzing GPU-controlled communication with dynamic parallelism in terms of performance and energy
Graphic Processing Units (GPUs) are widely used in high performance computing, due to their high computational power and high performance per Watt. However, one of the main bottlenecks of GPU-accelerated cluster computing is the data transfer between distributed GPUs. This not only affects performan...
Gespeichert in:
| Hauptverfasser: | , , |
|---|---|
| Dokumenttyp: | Article (Journal) |
| Sprache: | Englisch |
| Veröffentlicht: |
29 March 2016
|
| In: |
Parallel computing
Year: 2016, Jahrgang: 57, Pages: 125-134 |
| ISSN: | 1872-7336 |
| DOI: | 10.1016/j.parco.2016.02.005 |
| Online-Zugang: | Verlag, Volltext: http://dx.doi.org/10.1016/j.parco.2016.02.005 Verlag, Volltext: http://www.sciencedirect.com/science/article/pii/S0167819116300011 
|
| Verfasserangaben: | Lena Oden, Benjamin Klenk, Holger Fröning |
MARC
| LEADER | 00000caa a2200000 c 4500 | ||
|---|---|---|---|
| 001 | 1560390727 | ||
| 003 | DE-627 | ||
| 005 | 20220813185924.0 | ||
| 007 | cr uuu---uuuuu | ||
| 008 | 170703s2016 xx |||||o 00| ||eng c | ||
| 024 | 7 | |a 10.1016/j.parco.2016.02.005 |2 doi | |
| 035 | |a (DE-627)1560390727 | ||
| 035 | |a (DE-576)490390722 | ||
| 035 | |a (DE-599)BSZ490390722 | ||
| 035 | |a (OCoLC)1340976501 | ||
| 040 | |a DE-627 |b ger |c DE-627 |e rda | ||
| 041 | |a eng | ||
| 084 | |a 28 |2 sdnb | ||
| 100 | 1 | |a Oden, Lena |e VerfasserIn |0 (DE-588)1069864110 |0 (DE-627)822870797 |0 (DE-576)429555911 |4 aut | |
| 245 | 1 | 0 | |a Analyzing GPU-controlled communication with dynamic parallelism in terms of performance and energy |c Lena Oden, Benjamin Klenk, Holger Fröning |
| 264 | 1 | |c 29 March 2016 | |
| 300 | |a 10 | ||
| 336 | |a Text |b txt |2 rdacontent | ||
| 337 | |a Computermedien |b c |2 rdamedia | ||
| 338 | |a Online-Ressource |b cr |2 rdacarrier | ||
| 500 | |a Gesehen am 03.07.2017 | ||
| 520 | |a Graphic Processing Units (GPUs) are widely used in high performance computing, due to their high computational power and high performance per Watt. However, one of the main bottlenecks of GPU-accelerated cluster computing is the data transfer between distributed GPUs. This not only affects performance, but also power consumption. The most common way to utilize a GPU cluster is a hybrid model, in which the GPU is used to accelerate the computation, while the CPU is responsible for the communication. This approach always requires a dedicated CPU thread, which consumes additional CPU cycles and therefore increases the power consumption of the complete application. In recent work we have shown that the GPU is able to control the communication independently of the CPU. However, there are several problems with GPU-controlled communication. The main problem is intra-GPU synchronization, since GPU blocks are non-preemptive. Therefore, the use of communication requests within a GPU can easily result in a deadlock. In this work we show how dynamic parallelism solves this problem. GPU-controlled communication in combination with dynamic parallelism allows keeping the control flow of multi-GPU applications on the GPU and bypassing the CPU completely. Using other in-kernel synchronization methods results in massive performance losses, due to the forced serialization of the GPU thread blocks. Although the performance of applications using GPU-controlled communication is still slightly worse than the performance of hybrid applications, we will show that performance per Watt increases by up to 10% while still using commodity hardware. | ||
| 650 | 4 | |a Communication | |
| 650 | 4 | |a Data transfer | |
| 650 | 4 | |a Dynamic parallelism | |
| 650 | 4 | |a Energy efficiency | |
| 650 | 4 | |a GPUs | |
| 650 | 4 | |a Infiniband | |
| 700 | 1 | |a Klenk, Benjamin |d 1989- |e VerfasserIn |0 (DE-588)1073185729 |0 (DE-627)828640025 |0 (DE-576)434708488 |4 aut | |
| 700 | 1 | |a Fröning, Holger |d 1976- |e VerfasserIn |0 (DE-588)133209466 |0 (DE-627)538678658 |0 (DE-576)299696189 |4 aut | |
| 773 | 0 | 8 | |i Enthalten in |t Parallel computing |d Amsterdam [u.a.] : North-Holland, Elsevier Science, 1984 |g 57(2016), Seite 125-134 |h Online-Ressource |w (DE-627)265784115 |w (DE-600)1466340-5 |w (DE-576)074890999 |x 1872-7336 |7 nnas |a Analyzing GPU-controlled communication with dynamic parallelism in terms of performance and energy |
| 773 | 1 | 8 | |g volume:57 |g year:2016 |g pages:125-134 |g extent:10 |a Analyzing GPU-controlled communication with dynamic parallelism in terms of performance and energy |
| 856 | 4 | 0 | |u http://dx.doi.org/10.1016/j.parco.2016.02.005 |x Verlag |x Resolving-System |3 Volltext |
| 856 | 4 | 0 | |u http://www.sciencedirect.com/science/article/pii/S0167819116300011 |x Verlag |3 Volltext |
| 951 | |a AR | ||
| 992 | |a 20170703 | ||
| 993 | |a Article | ||
| 994 | |a 2016 | ||
| 998 | |g 133209466 |a Fröning, Holger |m 133209466:Fröning, Holger |d 700000 |d 720000 |e 700000PF133209466 |e 720000PF133209466 |k 0/700000/ |k 1/700000/720000/ |p 3 |y j | ||
| 998 | |g 1073185729 |a Klenk, Benjamin |m 1073185729:Klenk, Benjamin |d 700000 |d 720000 |e 700000PK1073185729 |e 720000PK1073185729 |k 0/700000/ |k 1/700000/720000/ |p 2 | ||
| 998 | |g 1069864110 |a Oden, Lena |m 1069864110:Oden, Lena |p 1 |x j | ||
| 999 | |a KXP-PPN1560390727 |e 2973050529 | ||
| BIB | |a Y | ||
| SER | |a journal | ||
| JSO | |a {"relHost":[{"origin":[{"publisher":"North-Holland, Elsevier Science ; North-Holland","dateIssuedKey":"1984","dateIssuedDisp":"1984-","publisherPlace":"Amsterdam [u.a.] ; Amsterdam [u.a.]"}],"id":{"issn":["1872-7336"],"eki":["265784115"],"zdb":["1466340-5"]},"note":["Gesehen am 31.05.23"],"language":["eng"],"physDesc":[{"extent":"Online-Ressource"}],"title":[{"title_sort":"Parallel computing","title":"Parallel computing"}],"recId":"265784115","type":{"bibl":"periodical","media":"Online-Ressource"},"pubHistory":["1.1984 - 40.2014; Vol. 41.2015 -"],"part":{"pages":"125-134","volume":"57","text":"57(2016), Seite 125-134","year":"2016","extent":"10"},"disp":"Analyzing GPU-controlled communication with dynamic parallelism in terms of performance and energyParallel computing"}],"type":{"bibl":"article-journal","media":"Online-Ressource"},"person":[{"given":"Lena","role":"aut","family":"Oden","display":"Oden, Lena"},{"family":"Klenk","display":"Klenk, Benjamin","role":"aut","given":"Benjamin"},{"given":"Holger","role":"aut","display":"Fröning, Holger","family":"Fröning"}],"recId":"1560390727","id":{"eki":["1560390727"],"doi":["10.1016/j.parco.2016.02.005"]},"origin":[{"dateIssuedKey":"2016","dateIssuedDisp":"29 March 2016"}],"name":{"displayForm":["Lena Oden, Benjamin Klenk, Holger Fröning"]},"note":["Gesehen am 03.07.2017"],"physDesc":[{"extent":"10 S."}],"language":["eng"],"title":[{"title":"Analyzing GPU-controlled communication with dynamic parallelism in terms of performance and energy","title_sort":"Analyzing GPU-controlled communication with dynamic parallelism in terms of performance and energy"}]} | ||
| SRT | |a ODENLENAKLANALYZINGG2920 | ||