MBS-LVM: A High-Performance Logical Volume Manager for Memory Bus-connected Storages over NUMA Servers
1
조회
0
다운로드
1
인용
0
참고문헌
키워드
초록
With the recent advances of memory technologies, high-performance non-volatile memories such as nonvolatiledual in-line memory module (NVDIMM) have begun to be used as an addition or an alternative toserver-side storages. When these memory bus-connected storages (MBSs) are installed over non-uniformmemory access (NUMA) servers, the distance between NUMA nodes and MBSs is one of the crucial factorsthat influence file processing performance, because the access latency of a NUMA system varies depending onits distance from the NUMA nodes. This paper presents the design and implementation of a high-performancelogical volume manager for MBSs, called MBS-LVM, when multiple MBSs are scattered over a NUMA server. The MBS-LVM consolidates the address space of each MBS into a single global address space and dynamicallyutilizes storage spaces such that each thread can access an MBS with the lowest latency possible. Weimplemented the MBS-LVM in the Linux kernel and evaluated its performance by porting it over the tmpfs, amemory-based file system widely used in Linux. The results of the benchmarking show that the writeperformance of the tmpfs using MBS-LVM has been improved by up to twenty times against the original tmpfsover a NUMA server with four nodes.
Abstract
With the recent advances of memory technologies, high-performance non-volatile memories such as nonvolatiledual in-line memory module (NVDIMM) have begun to be used as an addition or an alternative toserver-side storages. When these memory bus-connected storages (MBSs) are installed over non-uniformmemory access (NUMA) servers, the distance between NUMA nodes and MBSs is one of the crucial factorsthat influence file processing performance, because the access latency of a NUMA system varies depending onits distance from the NUMA nodes. This paper presents the design and implementation of a high-performancelogical volume manager for MBSs, called MBS-LVM, when multiple MBSs are scattered over a NUMA server. The MBS-LVM consolidates the address space of each MBS into a single global address space and dynamicallyutilizes storage spaces such that each thread can access an MBS with the lowest latency possible. Weimplemented the MBS-LVM in the Linux kernel and evaluated its performance by porting it over the tmpfs, amemory-based file system widely used in Linux. The results of the benchmarking show that the writeperformance of the tmpfs using MBS-LVM has been improved by up to twenty times against the original tmpfsover a NUMA server with four nodes.
참고문헌 (7)
- John Paul Martin, "Exploring the support for high performance applications in the container runtime environment", Human-centric Computing and Information Sciences, 8(1) (2018), Springer Science and Business Media LLC[DOI]
- Eren Gultepe, "Improving clustering performance using independent component analysis and unsupervised feature learning", Human-centric Computing and Information Sciences, 8(1) (2018), Springer Science and Business Media LLC[DOI]
- Alexey G. Finogeev, "The convergence computing model for big sensor data mining and knowledge discovery", Human-centric Computing and Information Sciences, 7(1) (2017), Springer Science and Business Media LLC[DOI]
- H. Cai, "IoT-based big data storage systems in cloud computing : perspectives and challenges", IEEE Internet of Things Journal, 4(1), pp.75~87 (2017)
- D. Niu, "XPMFS : a new NVM file system for vehicle big data", IEEE Access, 6, pp.34863~34873 (2018)
- J. Xu, "NOVA: a log-structured file system for hybrid volatile", non-volatile main memories, 2016, 323~338
- J. W. Kim, "ZonFS: a storage class memory file system with memory zone partitioning on Linux", Proceedings of 2017 IEEE 2nd International Workshops on Foundations and Applications of Self* Systems (FAS*W), pp.277~282 (2017)