Paper Info
Title
Accelerating by Idling: How Speculative Delays Improve Performance of Message-Oriented Systems.
Abstract
We propose a technique called speculative lagging, which improves performance by dynamically adding periods of idle execution into the message-oriented system. The speculation is guided by a statistical model, which predicts context switches that benefit from delays. We analytically derive the expected speedup, which, for a fixed confidence, allows identifying lagging opportunities in O(1) time, without a performance overhead. We describe the corresponding speculation algorithm and use it to extend an existing scheduler. Comparison with other actor frameworks on standard benchmarks shows improvements of up to 2.1(times ).
Year
Venue
Field
2017
Euro-Par
Speculation,Idle,Computer science,Parallel computing,Statistical model,Lagging,Context switch,Speedup
DocType
Citations 
PageRank 
Conference
1
0.35
References 
Authors
8
1
Name
Order
Citations
PageRank
Aleksandar Prokopec116313.56