Docstoc

Lithium_ Virtual Machine Storage for the Cloud

Document Sample
Lithium_ Virtual Machine Storage for the Cloud Powered By Docstoc
					Lithium: Virtual Machine
  Storage for the Cloud
               Jacob Gorm Hansen
                         VMware

                           Eric Jul
                 Bell Laboratories
                   Outline
•   Authors
•   Motivation & Background
•   Design & Implementation
•   Evaluation
•   Related work
                       Authors
• Jacob Gorm Hansen
      @VMware, Aarhus, Denmark
      http://www.diku.dk/~jacobg/
• Eric Jul
      @Bell Laboratories, Alcatel-Lucent, Dublin, Ireland
      http://www.diku.dk/~eric/
                 Motivation
• Address the limitations of centralized shared
  storage for cloud computing
• The old systems trade backwards-
  compatibility in exchange for better scalability
  and resiliency to disk and host failures.
• a number of scalable storage systems are do
  not fit the requirements of VM storage
Design
                              Design
• Fork-consistent Replication
      • Protect integrity and consistency of replicas
• Replica Consistency
      • Stateless and supports an unlimited amount of both lagging and
        synchronous replicas.
• Log-structuring
      •   Temporal Order Preservation.
      •   Optimized for Random Writes.
      •   Live Storage Migration.
      •   Flash-compatible.
• Mutual Exclusion
      • Access Token
      • Automated Fail-over
     Fork-consistent Replication
• A partially ordered data model
• Protect the integrity and consistency of
  replicas.
• partial ordering and strong checksums
           Replica Consistency
• two-phase commit (2PC) protocol (having to
  write everything twice)
• dirty-bitmap(requires a bitmap per replica)
• Lithium's replication protocol is stateless and
  supports an unlimited amount of both lagging
  and synchronous replicas.
              Log-structuring
•   Temporal Order Preservation
•   Optimized for Random Writes
•   Live Storage Migration
•   Flash-compatible
            Mutual Exclusion
• An extension to the fork-consistency protocol
• Access Token
• Automated Fail-over
               Implementation
• B-tree Indexes
     • Does not track individual block locations, but entire
       extents.
• Log Compaction
• Locating Updates on Disk
     • Use a Bloom filter that describes the contents of each
       log segment on the disk.
• POSIX Interface
     • Emulating POSIX introduces scalability bottlenecks.
     • Provide a per-volume POSIX layer on top of Lithium.
                    Evaluation
• Tools
     • PostMark.
     • DVDStore2.
• Replication Performance
• Scalability
Replication Performance
Scalability
                Related work
•   The Google File System
•   Boxwood
•   Parallax
•   The HP Federated Array of Bricks
•   Sun's Zettabyte File System
                Conclusion
• Lithium is able to compete with an expensive
  Fibre Channel storage array on a small cluster
  of eight hosts, and is faster than traditional
  disk layouts for replication workloads.
• https://github.com/vmware/CloudFS

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:0
posted:6/18/2013
language:English
pages:16