Using Deduplicating Storage for Efficient Disk Image Deployment Xing Lin, Mike Hibler, Eric Eide, Robert Ricci University of Utah.

Slides:



Advertisements
Similar presentations
<Insert Picture Here>
Advertisements

Consistency and Replication Chapter 7 Part II Replica Management & Consistency Protocols.
Yuchong Hu1, Henry C. H. Chen1, Patrick P. C. Lee1, Yang Tang2
1 Securing the Frisbee Multicast Disk Loader Robert Ricci, Jonathon Duerig University of Utah.
File Systems.
Esma Yildirim Department of Computer Engineering Fatih University Istanbul, Turkey DATACLOUD 2013.
11-May-15CSE 542: Operating Systems1 File system trace papers The Zebra striped network file system. Hartman, J. H. and Ousterhout, J. K. SOSP '93. (ACM.
CacheCast: Eliminating Redundant Link Traffic for Single Source Multiple Destination Transfers Piotr Srebrny, Thomas Plagemann, Vera Goebel Department.
Transparent Checkpoint of Closed Distributed Systems in Emulab Anton Burtsev, Prashanth Radhakrishnan, Mike Hibler, and Jay Lepreau University of Utah,
Low-Cost Data Deduplication for Virtual Machine Backup in Cloud Storage Wei Zhang, Tao Yang, Gautham Narayanasamy University of California at Santa Barbara.
1 Live Deduplication Storage of Virtual Machine Images in an Open-Source Cloud Chun-Ho Ng, Mingcao Ma, Tsz-Yeung Wong, Patrick P. C. Lee, John C. S. Lui.
Video Staging: A Proxy-Server- Based Approach to End-to-End Video Delivery over Wide-Area Networks Zhi-Li Zhang, Yuewei Wang, David H.C Du, Dongli Su Άννα.
An Adaptable Benchmark for MPFS Performance Testing A Master Thesis Presentation Yubing Wang Advisor: Prof. Mark Claypool.
G Robert Grimm New York University SGI’s XFS or Cool Pet Tricks with B+ Trees.
1 Fast, Scalable Disk Imaging with Frisbee University of Utah Mike Hibler, Leigh Stoller, Jay Lepreau, Robert Ricci, Chad Barb.
Integrated Scientific Workflow Management for the Emulab Network Testbed Eric Eide, Leigh Stoller, Tim Stack, Juliana Freire, and Jay Lepreau and Jay Lepreau.
Exploiting SCI in the MultiOS management system Ronan Cunniffe Brian Coghlan SCIEurope’ AUG-2000.
1 stdchk : A Checkpoint Storage System for Desktop Grid Computing Matei Ripeanu – UBC Sudharshan S. Vazhkudai – ORNL Abdullah Gharaibeh – UBC The University.
U NIVERSITY OF M ASSACHUSETTS, A MHERST Department of Computer Science Emery Berger University of Massachusetts Amherst Operating Systems CMPSCI 377 Lecture.
Multi-level Selective Deduplication for VM Snapshots in Cloud Storage Wei Zhang*, Hong Tang †, Hao Jiang †, Tao Yang*, Xiaogang Li †, Yue Zeng † * University.
RAID-x: A New Distributed Disk Array for I/O-Centric Cluster Computing Kai Hwang, Hai Jin, and Roy Ho.
1 Proxy-Assisted Techniques for Delivering Continuous Multimedia Streams Lixin Gao, Zhi-Li Zhang, and Don Towsley.
Slingshot: Deploying Stateful Services in Wireless Hotspots Ya-Yunn Su Jason Flinn University of Michigan.
Data Deduplication in Virtualized Environments Marc Crespi, ExaGrid Systems
ObliviStore High Performance Oblivious Cloud Storage Emil StefanovElaine Shi
RAID COP 5611 Advanced Operating Systems Adapted from Andy Wang’s slides at FSU.
Lecture 9 of Advanced Databases Storage and File Structure (Part II) Instructor: Mr.Ahmed Al Astal.
Review of Memory Management, Virtual Memory CS448.
Min Xu1, Yunfeng Zhu2, Patrick P. C. Lee1, Yinlong Xu2
Profiling Grid Data Transfer Protocols and Servers George Kola, Tevfik Kosar and Miron Livny University of Wisconsin-Madison USA.
Improving Disk Latency and Throughput with VMware Presented by Raxco Software, Inc. March 11, 2011.
Oracle Advanced Compression – Reduce Storage, Reduce Costs, Increase Performance Session: S Gregg Christman -- Senior Product Manager Vineet Marwah.
File System Implementation Chapter 12. File system Organization Application programs Application programs Logical file system Logical file system manages.
Improving Content Addressable Storage For Databases Conference on Reliable Awesome Projects (no acronyms please) Advanced Operating Systems (CS736) Brandon.
Amy Apon, Pawel Wolinski, Dennis Reed Greg Amerson, Prathima Gorjala University of Arkansas Commercial Applications of High Performance Computing Massive.
A Measurement Based Memory Performance Evaluation of High Throughput Servers Garba Isa Yau Department of Computer Engineering King Fahd University of Petroleum.
Fragmentation in Large Object Repositories Russell Sears Catharine van Ingen CIDR 2007 This work was performed at Microsoft Research San Francisco with.
Kiew-Hong Chua a.k.a Francis Computer Network Presentation 12/5/00.
1 CloudVS: Enabling Version Control for Virtual Machines in an Open- Source Cloud under Commodity Settings Chung-Pan Tang, Tsz-Yeung Wong, Patrick P. C.
1 Evaluation of Cooperative Web Caching with Web Polygraph Ping Du and Jaspal Subhlok Department of Computer Science University of Houston presented at.
1 Public DAFS Storage for High Performance Computing using MPI-I/O: Design and Experience Arkady Kanevsky & Peter Corbett Network Appliance Vijay Velusamy.
ROOT and Federated Data Stores What Features We Would Like Fons Rademakers CERN CC-IN2P3, Nov, 2011, Lyon, France.
RevDedup: A Reverse Deduplication Storage System Optimized for Reads to Latest Backups Chun-Ho Ng, Patrick P. C. Lee The Chinese University of Hong Kong.
Latency Reduction Techniques for Remote Memory Access in ANEMONE Mark Lewandowski Department of Computer Science Florida State University.
Large-scale Virtualization in the Emulab Network Testbed Mike Hibler, Robert Ricci, Leigh Stoller Jonathon Duerig Shashi Guruprasad, Tim Stack, Kirk Webb,
Improving Disk Throughput in Data-Intensive Servers Enrique V. Carrera and Ricardo Bianchini Department of Computer Science Rutgers University.
FAT File Allocation Table
Enabling Grids for E-sciencE EGEE and gLite are registered trademarks Tools and techniques for managing virtual machine images Andreas.
Review CS File Systems - Partitions What is a hard disk partition?
 2004 Deitel & Associates, Inc. All rights reserved. Chapter 9 – Real Memory Organization and Management Outline 9.1 Introduction 9.2Memory Organization.
1 Evaluation of Cooperative Web Caching with Web Polygraph Ping Du and Jaspal Subhlok Department of Computer Science University of Houston presented at.
October 15-18, 2013 Charlotte, NC Accelerating Database Performance Using Compression Joseph D’Antoni, Solutions Architect Anexinet.
Performance Evaluation of Redirection Schemes in Content Distribution Networks Jussi Kangasharju, Keith W. Ross Institut Eurecom Jim W. Roberts France.
Extending Auto-Tiering to the Cloud For additional, on-demand, offsite storage resources 1.
Presenter: Yue Zhu, Linghan Zhang A Novel Approach to Improving the Efficiency of Storing and Accessing Small Files on Hadoop: a Case Study by PowerPoint.
Canadian Bioinformatics Workshops
PHD Virtual Technologies “Reader’s Choice” Preferred product.
RAID Overview.
Demystifying Deduplication
Chapter 9 – Real Memory Organization and Management
CSI 400/500 Operating Systems Spring 2009
Securing the Frisbee Multicast Disk Loader
The Google File System Sanjay Ghemawat, Howard Gobioff and Shun-Tak Leung Google Presented by Jiamin Huang EECS 582 – W16.
Database Implementation Issues
So far… Text RO …. printf() RW link printf Linking, loading
Cooperative Caching, Simplified
Declarative Transfer Learning from Deep CNNs at Scale
Jingwei Li*, Patrick P. C. Lee #, Yanjing Ren*, and Xiaosong Zhang*
Database Implementation Issues
Fan Ni Xing Lin Song Jiang
Presentation transcript:

Using Deduplicating Storage for Efficient Disk Image Deployment Xing Lin, Mike Hibler, Eric Eide, Robert Ricci University of Utah

this talk utilizing a deduplicating storage system within a fast disk-imaging system 3× decrease in storage negligible run-time overhead “don’t be the bottleneck” Aligned Fixed-size Chunking 2 VF results techniques

3 disk image server loaded on demand be fast! deliver data as fast as clients can receive it

4 disk image server

5 Utah Emulab 1,000+ disk images 21 TB total Amazon EC2 37,000+ public AMIs fast ☛ ☚ compact

deduplication 6 dedup. storage system

deduplication 7 Image 1 fingerprint 1; fingerprint 2; fingerprint 3; … Image 1 fingerprint 1; fingerprint 2; fingerprint 3; … Image 2 fingerprint 1; fingerprint 2; fingerprint 19; … Image 2 fingerprint 1; fingerprint 2; fingerprint 19; … small “recipe”

dedup. for disk images images are often derived from other images – users add packages to testbed “base” images – users’ work-in-progress snapshots – … a lot of duplicated data across images! 8

9 disk image server

10 disk image server dedup. disk image storage problem: dedup. storage can be slow our contrib: add dedup. without system slowdown

why is frisbee fast? compression use filesystem info pipeline independent “chunks” 11 disk image server lower network bandwidth smaller files fewer disk writes disk read net xfer decompress disk write keep receiving disk busy keep pipeline filled new clients can join sequential disk writes

from frisbee to VF Frisbee: disk images stored as files VF: disk image data stored in Venti reformed into chunks by Chunkmaker 12 disk image server Venti Chunkmaker [Quinlan & Dorward, FAST ’02] [Rhea et al., ATC ’08]

image corpus 430 Linux images from Utah Emulab – 76 “standard” images – 354 user-created images based on RedHat, Fedora, CentOS, & Ubuntu 13 Venti

addressing the challenges compression use filesystem info pipeline independent “chunks”

compression 15 Venti image server capture partition store retrieve

compression 16 Venti image server capture partition store retrieve compressed disk image compress poor deduplication (1.11×)

compression 17 Venti image server capture partition store retrieve

compression 18 Venti image server capture partition store retrieve compress disk data compress too slow compress30.29 MB/s disk write71.07 MB/s

compression 19 Venti image server capture partition store retrieve

compression 20 Venti image server capture partition retrieve store compressed dedup blocks compress preserves opportunities for dedup server retrieves & concatenates compressed blocks to form chunks 6% more chunks vs. original Frisbee

addressing the challenges compression use filesystem info pipeline independent “chunks”

use filesystem info exclude unallocated sectors from image promote sequential disk writes process the “stream” of allocated sectors 22

23 Venti sector stream make dedup blocks via “fixed-size chunking”

24 Venti sector allocations & frees move the dedup block boundaries fixed-sized chunking over sector stream leads to poor deduplication across disk images abc abc abc345678

aligned fixed-size chunking abc deduplicate! Venti 12zz z zzab czzz block boundaries based on sector offsets “pad” partially filled blocks with zero sectors

how big should dedup blocks be? better dedup – more likely to match slower – more accesses to Venti lower compression ratio – less data per block more metadata per image lower dedup – less likely to match faster – fewer accesses to Venti higher compression ratio – more data per block less metadata per image 26 big — say, 48K small — say, 4K

addressing the challenges compression use filesystem info pipeline independent “chunks”

pipeline speed through parallelism choose maximum storage benefit that doesn’t slow down the pipeline 28 disk image server Venti read net xfer decompress disk write Venti i.e., the smallest dedup block size

29 ✖✖✖✔

image 32K (compressed) image data: GB (compressed) data in Venti: GB deduplication ratio: 3.26 image metadata: GB total space savings versus Frisbee: 67.8% 30

addressing the challenges compression use filesystem info pipeline independent “chunks”

independent chunks chunk — Frisbee’s network protocol unit – contains multiple groups of sectors – client requests chunks until it has them all 32 disk image server Venti Chunkmaker Metadata chunk headers; fingerprints Metadata chunk headers; fingerprints

independent chunks client requests chunk find precomputed chunk metadata – chunk header – dedup block fingerprints retrieve dedup blocks from Venti concatenate blocks with header and transmit to client cache constructed chunk 33 disk image server Venti Chunkmaker Metadata chunk headers; fingerprints Metadata chunk headers; fingerprints

evaluation storage savings synchronized deployment staggered deployment

storage savings load our image corpus into Venti – 430 Linux images – load from oldest to newest track storage as images are added – compressed, dedup’ed data in Venti – storage required by “baseline Frisbee” 35

36 3× 233 GB 75 GB

disk image deployment setup 1 Gbps switched LAN single server – running “baseline Frisbee” or VF – configured to distribute data at 500 Mbps up to 20 client machines – Dell PowerEdge R710s (see paper for specs) 37

synchronized deployment deploy single disk image to – 1 client – 8 clients that start at the same time – 16 clients that start at the same time measure time to deploy over 10 trials (image: 1.4 GB uncompressed data) 38

39 2% increase in run time

staggered deployment deploy single disk image to – 20 clients – organized into 5 groups – groups start at 5-second intervals measure time to deploy over 10 trials 40

41 3% increase in run time

conclusions VF combines deduplicating storage with a high-performance disk distribution system 3× reduction in required storage 2–3% run-time overhead “don’t be the bottleneck”: careful design – obtain dedup benefit: AFC – preserve existing optimizations 42