50 likes | 192 Vues
This document provides an overview of the pNFS Block Layout for EMC Linux systems, focusing on the improved performance and transparency to applications when interfacing with different storage vendors. It discusses the role of the pNFS Client Layout Driver and iSCSI Protocol in normalizing access to clustered file systems. Emphasis is placed on the enhancements introduced in the latest EMC VNX release, which improves block performance, particularly for small files. Key performance metrics are highlighted, demonstrating significant scalability with tested configurations.
E N D
Linux pNFS Block Layout Transparent to applications Common client for different storage vendors Fewer support issues for storage vendors Normalizes access to clustered file systems Client Apps • blocks (iSCSI) NFSv4.1 Layoutgrant & recall pNFS Client Layout Driver pNFS Server iSCSI Protocol Cluster File System Control Protocol
NFSv4.1 – with pNFS Block Linux pNFS block layout Client Support generic layout – CITI Univ. of Michigan Support block layout client (iSCSI) - F-16 CITI Maintenance of block layout – CITI via Linux kernel Bugzilla Performance monitoring and patching – CITI via Bugzilla New EMC VNX release with improved pNFS block performance – Q3 2012 High bandwidth Improved small files performance Support to RedHat Support block layout – EMC Elabwork with RedHat to qualify pNFS EMC Elab will qualify Fedora 16 3
pNFS Block – Protocol enhancement • NFSv4.1/pNFS were standardized at IETF • NFSv4 working group (WG) • New Disk Protection draft approved by IESG • Extension to RFC 5663 • Added mechanism that enables identification of block storage devices used by pNFS to be protected using GPT • EMC VNX will implement it when RFC is approved
pNFS Block Performance Latest Performance Using 10GbE server iSCSI and 1 GbE clients 4 threads per client Single Client Write: 107 MB/sec Read: 90 MB/sec 8 Clients Write: 852 MB/sec – 98% scalability Read: 674 MB/sec – 96% scalability 5