Title
Per-VM page cache partitioning for cloud computing platforms
Abstract
Cloud computing has become popular for shared hosting of third-party applications. A cloud platform may multiplex virtual machines running different customer applications onto a single physical server, raising the potential for performance interference between such applications. In particular, when a hypervisor shares the file system page cache between virtual machines, as is common in Linux environments, it is possible for one VM to impact the performance seen by other co-located VMs. To address this drawback and improve performance isolation, we design a page cache which is partitioned by VMs. Such a design provides the ability to control fine-grained caching parameters such as cache size and eviction policies individually. Furthermore, the deterministic cache allocation and partitioning provides improved performance isolation among VMs. We provide dynamic cache partitioning by using utility derived from the miss-ratio characteristics. We implement our page cache architecture in the Linux kernel and demonstrate its efficacy using disk image files of virtual machines and different types of file access patterns by applications. Experimental results show that the utility-based partitioning can reduce the cache size by up to an order of magnitude while increasing cache hit ratios by up to 20%. Among other features, the per-file page cache has fadvise integration, a scan-resistant eviction algorithm (ARC) and reduced lock-contention and overhead during the eviction process.
Year
DOI
Venue
2016
10.1109/COMSNETS.2016.7439971
International Conference on Communication Systems and Networks
Field
DocType
ISSN
Cache-oblivious algorithm,Cache invalidation,Cache pollution,Cache,Computer science,Computer network,Page cache,Cache algorithms,Real-time computing,Cache coloring,Smart Cache,Operating system
Conference
2155-2487
Citations 
PageRank 
References 
3
0.38
22
Authors
3
Name
Order
Citations
PageRank
Prateek Sharma120114.12
Purushottam Kulkarni2130667.96
Prashant J. Shenoy36386521.30