2010 31st IEEE Real-Time Systems Symposium 2010
DOI: 10.1109/rtss.2010.23
|View full text |Cite
|
Sign up to set email alerts
|

An Empirical Comparison of Global, Partitioned, and Clustered Multiprocessor EDF Schedulers

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
1
1
1
1

Citation Types

2
102
0

Year Published

2012
2012
2021
2021

Publication Types

Select...
4
3
2

Relationship

1
8

Authors

Journals

citations
Cited by 117 publications
(104 citation statements)
references
References 15 publications
2
102
0
Order By: Relevance
“…Interestingly, the aTC-allocator success ratio appears to be quite close to that of DSE, which allows concluding that our algorithm is sufficiently good at finding feasible partitions. Notably, the success ratio decreases when the total utilization in isolation reaches 2.4 − 2.5 out of 4, which is consistent with state-ofthe-art results in the analysis of schedulable utilization with partitioned scheduling [6]. In the next section we analyze in detail why aTC-allocator does not reach good success rates for higher utilizations.…”
Section: Results With Atc Etbsupporting
confidence: 83%
“…Interestingly, the aTC-allocator success ratio appears to be quite close to that of DSE, which allows concluding that our algorithm is sufficiently good at finding feasible partitions. Notably, the success ratio decreases when the total utilization in isolation reaches 2.4 − 2.5 out of 4, which is consistent with state-ofthe-art results in the analysis of schedulable utilization with partitioned scheduling [6]. In the next section we analyze in detail why aTC-allocator does not reach good success rates for higher utilizations.…”
Section: Results With Atc Etbsupporting
confidence: 83%
“…We can assume that this gap in system load between the two strategies will increase with more processors. This is in accordance with the results of Bastoni et al stating that G-EDF is not a viable choice for hard real-time systems with a large number of processors (24 in their study) (Bastoni et al, 2010). The computation times of the jobs are shorter with P-EDF compared to G-EDF as shown in Figure 7.…”
Section: Application Examplesupporting
confidence: 91%
“…For any time t > 0, the notation t − is used to denote the time t − ε in the limit ε → 0+, and the notation t + is used to denote the time t + ε in the limit ε → 0+. 2 The SRT analysis framework used here has been adopted from a framework for ordinary sporadic task systems first proposed in [7], and subsequently used in several other papers [15][16][17]. …”
Section: Response Time Boundmentioning
confidence: 99%
“…4 Each such parallel task set was generated by creating parallel tasks until total utilization exceeded U sum , and by then reducing the last task's utilization so that the total system utilization equalled U sum . For each generated system, we first checked schedulability (i.e., the ability to ensure bounded 4 For systems with higher processor counts, recent experimental work [2] suggests that when overheads are considered, clustered scheduling approaches (where groups of processors with low processor counts that share low-level caches are scheduled globally) are better than global approaches. response times) and the magnitude of response time bounds using Theorem 1.…”
Section: Experimental Evaluationmentioning
confidence: 99%