Details
| Originalsprache | Englisch |
|---|---|
| Titel des Sammelwerks | INFOCOM 2020 |
| Untertitel | IEEE Conference on Computer Communications |
| Herausgeber (Verlag) | Institute of Electrical and Electronics Engineers Inc. |
| Seiten | 1063-1072 |
| Seitenumfang | 10 |
| ISBN (elektronisch) | 9781728164120 |
| ISBN (Print) | 978-1-7281-6413-7 |
| Publikationsstatus | Veröffentlicht - 2020 |
| Veranstaltung | 38th IEEE Conference on Computer Communications, INFOCOM 2020 - Toronto, Kanada Dauer: 6 Juli 2020 → 9 Juli 2020 |
Publikationsreihe
| Name | Proceedings - IEEE INFOCOM |
|---|---|
| Band | 2020-July |
| ISSN (Print) | 0743-166X |
Abstract
Models of parallel processing systems typically assume that one has l servers and jobs are split into an equal number of k = l tasks. This seemingly simple approximation has surprisingly large consequences for the resulting stability and performance bounds. In reality, best practices for modern mapreduce systems indicate that a job's partitioning factor should be much larger than the number of servers available, with some researchers going to far as to advocate for a tiny tasks regime, where jobs are split into over 10,000 tasks. In this paper we use recent advances in stochastic network calculus to fundamentally understand the effects of task granularity on parallel systems' scaling, stability, and performance. For the split-merge model, we show that when one allows for tiny tasks, the stability region is actually much better than had previously been concluded. For the single-queue fork-join model, we show that sojourn times quickly approach the optimal case when l big tasks are subdivided into k≫ l tiny tasks. Our results are validated using extensive simulations, and the applicability of the models used is validated by experiments on an Apache Spark cluster.
ASJC Scopus Sachgebiete
- Informatik (insg.)
- Allgemeine Computerwissenschaft
- Ingenieurwesen (insg.)
- Elektrotechnik und Elektronik
Zitieren
- Standard
- Harvard
- Apa
- Vancouver
- BibTex
- RIS
INFOCOM 2020 : IEEE Conference on Computer Communications. Institute of Electrical and Electronics Engineers Inc., 2020. S. 1063-1072 9155368 (Proceedings - IEEE INFOCOM; Band 2020-July).
Publikation: Beitrag in Buch/Bericht/Sammelwerk/Konferenzband › Aufsatz in Konferenzband › Forschung › Peer-Review
}
TY - GEN
T1 - Tiny Tasks
T2 - 38th IEEE Conference on Computer Communications, INFOCOM 2020
AU - Fidler, Markus
AU - Walker, Brenton
AU - Bora, Stefan
N1 - Funding information: This work was supported in part by the German Research Council (DFG) under Grant VaMoS (FI 1236/7-1).
PY - 2020
Y1 - 2020
N2 - Models of parallel processing systems typically assume that one has l servers and jobs are split into an equal number of k = l tasks. This seemingly simple approximation has surprisingly large consequences for the resulting stability and performance bounds. In reality, best practices for modern mapreduce systems indicate that a job's partitioning factor should be much larger than the number of servers available, with some researchers going to far as to advocate for a tiny tasks regime, where jobs are split into over 10,000 tasks. In this paper we use recent advances in stochastic network calculus to fundamentally understand the effects of task granularity on parallel systems' scaling, stability, and performance. For the split-merge model, we show that when one allows for tiny tasks, the stability region is actually much better than had previously been concluded. For the single-queue fork-join model, we show that sojourn times quickly approach the optimal case when l big tasks are subdivided into k≫ l tiny tasks. Our results are validated using extensive simulations, and the applicability of the models used is validated by experiments on an Apache Spark cluster.
AB - Models of parallel processing systems typically assume that one has l servers and jobs are split into an equal number of k = l tasks. This seemingly simple approximation has surprisingly large consequences for the resulting stability and performance bounds. In reality, best practices for modern mapreduce systems indicate that a job's partitioning factor should be much larger than the number of servers available, with some researchers going to far as to advocate for a tiny tasks regime, where jobs are split into over 10,000 tasks. In this paper we use recent advances in stochastic network calculus to fundamentally understand the effects of task granularity on parallel systems' scaling, stability, and performance. For the split-merge model, we show that when one allows for tiny tasks, the stability region is actually much better than had previously been concluded. For the single-queue fork-join model, we show that sojourn times quickly approach the optimal case when l big tasks are subdivided into k≫ l tiny tasks. Our results are validated using extensive simulations, and the applicability of the models used is validated by experiments on an Apache Spark cluster.
UR - http://www.scopus.com/inward/record.url?scp=85090297693&partnerID=8YFLogxK
U2 - 10.1109/infocom41043.2020.9155368
DO - 10.1109/infocom41043.2020.9155368
M3 - Conference contribution
AN - SCOPUS:85090297693
SN - 978-1-7281-6413-7
T3 - Proceedings - IEEE INFOCOM
SP - 1063
EP - 1072
BT - INFOCOM 2020
PB - Institute of Electrical and Electronics Engineers Inc.
Y2 - 6 July 2020 through 9 July 2020
ER -