Paper Info
Title
Multilayer cache partitioning for multiprogram workloads
Abstract
We present a fully-automated, model based, multilayer cache partitioning scheme for multiprogram workloads running on multicore machines. As opposed to prior efforts, this scheme partitions shared caches at multiple layers simultaneously in a coordinated fashion. This scheme tries to achieve two objectives. First, it tries to satisfy the specified quality of service (QoS) values for all applications by partitioning the shared cache hierarchy across them, and second, it distributes the remaining excess cache capacity (if any) across applications such that a global performance metric is maximized. Our experimental analysis shows that the proposed multilayer partitioning scheme generates, on average, 33.1% improvement (on the weighted speedup metric) over the next best-performing scheme and is very successful in satisfying the QoS requirements of applications. Also, we show that partitioning each layer in isolation cannot generate the benefits obtained through our coordinated partitioning scheme. In addition, we observed that the difference between our scheme and an optimal scheme (that derives best dynamic partitions) was less than 15% for all the workloads tested and 6.6% on average.
Year
DOI
Venue
2011
10.1007/978-3-642-23400-2_13
Euro-Par (1)
Keywords
Field
DocType
remaining excess cache capacity,optimal scheme,partitioning scheme,global performance metric,shared cache hierarchy,multilayer cache,proposed multilayer partitioning scheme,multiprogram workloads,next best-performing scheme,qos requirement
Shared memory,Cache capacity,Computer science,Cache,Performance metric,Parallel computing,Quality of service,Hierarchy,Multi-core processor,Distributed computing,Speedup
Conference
Volume
ISSN
Citations 
6852
0302-9743
2
PageRank 
References 
Authors
0.35
14
4
Name
Order
Citations
PageRank
Mahmut T. Kandemir17371568.54
Ramya Prabhakar2747.27
Mustafa Karakoy3374.11
Yuanrui Zhang418015.48