Rim Moussa University Paris 9 Dauphine Experimental Performance Analysis of LH* RS Parity Management Workshop on Distributed Data Structures: WDAS 2002.

Slides:



Advertisements
Similar presentations
Network Resource Broker for IPTV in Cloud Computing Lei Liang, Dan He University of Surrey, UK OGF 27, G2C Workshop 15 Oct 2009 Banff,
Advertisements

Digital Library Service – An overview Introduction System Architecture Components and their functionalities Experimental Results.
1 Interoperability of a Scalable Distributed Data Manager with an Object-relational DBMS Thesis presentation Yakham NDIAYE November, 13 the 2001 November,
Design & Implementation of LH* RS : a Highly- Available Distributed Data Structure Rim Moussa
Contribution to the Design & Implementation of the Highly Available Scalable and Distributed Data Structure: LH* RS Rim Moussa
ZHT 1 Tonglin Li. Acknowledgements I’d like to thank Dr. Ioan Raicu for his support and advising, and the help from Raman Verma, Xi Duan, and Hui Jin.
The TickerTAIP Parallel RAID Architecture P. Cao, S. B. Lim S. Venkatraman, J. Wilkes HP Labs.
Serverless Network File Systems. Network File Systems Allow sharing among independent file systems in a transparent manner Mounting a remote directory.
A Server-less Architecture for Building Scalable, Reliable, and Cost-Effective Video-on-demand Systems Jack Lee Yiu-bun, Raymond Leung Wai Tak Department.
Cooperative Caching of Dynamic Content on a Distributed Web Server Vegard Holmedahl, Ben Smith, Tao Yang Speaker: SeungLak Choi, DB Lab., CS Dept.
Extensible Networking Platform IWAN 2005 Extensible Network Configuration and Communication Framework Todd Sproull and John Lockwood
1 Principles of Reliable Distributed Systems Tutorial 12: Frangipani Spring 2009 Alex Shraer.
Lecture 6 – Google File System (GFS) CSE 490h – Introduction to Distributed Computing, Winter 2008 Except as otherwise noted, the content of this presentation.
Scalable and Distributed Similarity Search in Metric Spaces Michal Batko Claudio Gennaro Pavel Zezula.
Object Naming & Content based Object Search 2/3/2003.
Witold Litwin Riad Mokadem Thomas Schwartz Disk Backup Through Algebraic Signatures.
September 24, 2007The 3 rd CSAIL Student Workshop Byzantine Fault Tolerant Cooperative Caching Raluca Ada Popa, James Cowling, Barbara Liskov Summer UROP.
Wide-area cooperative storage with CFS
WDAS Workshop, Lausanne, Jul. 9th1 Implementing SD-SQL Server: a Scalable Distributed Database System Soror SAHRI Witold LITWIN
Low Latency Wireless Video Over Networks Using Path Diversity John Apostolopolous Wai-tian Tan Mitchell Trott Hewlett-Packard Laboratories Allen.
Locality-Aware Request Distribution in Cluster-based Network Servers Presented by: Kevin Boos Authors: Vivek S. Pai, Mohit Aron, et al. Rice University.
MetaSync File Synchronization Across Multiple Untrusted Storage Services Seungyeop Han Haichen Shen, Taesoo Kim*, Arvind Krishnamurthy,
Frangipani: A Scalable Distributed File System C. A. Thekkath, T. Mann, and E. K. Lee Systems Research Center Digital Equipment Corporation.
Distributed Databases Dr. Lee By Alex Genadinik. Distributed Databases? What is that!?? Distributed Database - a collection of multiple logically interrelated.
Sanjay Ghemawat, Howard Gobioff, and Shun-Tak Leung Google∗
Cloud MapReduce : a MapReduce Implementation on top of a Cloud Operating System Speaker : 童耀民 MA1G Authors: Huan Liu, Dan Orban Accenture.
1 The Google File System Reporter: You-Wei Zhang.
©2001 Pål HalvorsenINFOCOM 2001, Anchorage, April 2001 Integrated Error Management in MoD Services Pål Halvorsen, Thomas Plagemann, and Vera Goebel University.
A brief overview about Distributed Systems Group A4 Chris Sun Bryan Maden Min Fang.
Database Design – Lecture 16
LH* RS P2P : A Scalable Distributed Data Structure for P2P Environment W. LITWIN CERIA Laboratory H.YAKOUBEN Paris Dauphine University
So, Jung-ki Distributed Computing System LAB School of Computer Science and Engineering Seoul National University Implementation of Package Management.
1 Configurable Security for Scavenged Storage Systems NetSysLab The University of British Columbia Abdullah Gharaibeh with: Samer Al-Kiswany, Matei Ripeanu.
Introduction to Hadoop and HDFS
1 High-Availability LH* Schemes with Mirroring W. Litwin, M.-A. Neimat U. Paris 9 & HPL Palo-Alto
Speaker: 吳晋賢 (Chin-Hsien Wu) Embedded Computing and Applications Lab Department of Electronic Engineering National Taiwan University of Science and Technology,
Distributed Session Announcement Agents for Real-time Streaming Applications Keio University, Graduate School of Media and Governance Kazuhiro Mishima.
Amy Apon, Pawel Wolinski, Dennis Reed Greg Amerson, Prathima Gorjala University of Arkansas Commercial Applications of High Performance Computing Massive.
CEPH: A SCALABLE, HIGH-PERFORMANCE DISTRIBUTED FILE SYSTEM S. A. Weil, S. A. Brandt, E. L. Miller D. D. E. Long, C. Maltzahn U. C. Santa Cruz OSDI 2006.
Scalable Distributed Data Structures & High-Performance Computing Witold Litwin Fethi Bennour CERIA University Paris 9 Dauphine
Presenters: Rezan Amiri Sahar Delroshan
1 High-Availability in Scalable Distributed Data Structures W. Litwin.
1 WDAS – 14 June THESSALONIKI(Greece) Range Queries to Scalable Distributed Data Structure RP* WDAS – 14 June THESSALONIKI(Greece) Range.
Fast Crash Recovery in RAMCloud. Motivation The role of DRAM has been increasing – Facebook used 150TB of DRAM For 200TB of disk storage However, there.
Providing Differentiated Levels of Service in Web Content Hosting Jussara Almeida, etc... First Workshop on Internet Server Performance, 1998 Computer.
Computer Science Lecture 19, page 1 CS677: Distributed OS Last Class: Fault tolerance Reliable communication –One-one communication –One-many communication.
1 Scalable Distributed Data Structures Part 2 Witold Litwin Paris 9
1 Xin Wang Internet Real -Time Laboratory Internet Real -Time Laboratory Columbia University ( Joint work with Henning Schulzrinne, Dilip Kandlur, and.
HADOOP DISTRIBUTED FILE SYSTEM HDFS Reliability Based on “The Hadoop Distributed File System” K. Shvachko et al., MSST 2010 Michael Tsitrin 26/05/13.
LH* RS P2P : A Scalable Distributed Data Structure for P2P Environment W. LITWIN CERIA Laboratory H.YAKOUBEN Paris Dauphine University
1 Distributed Processing Chapter 1 : Introduction.
EFFICIENT SOLUTION TO REPLICATED LOG AND DICTIONARY PROBLEM. (Gene T.J. Wuu & Arthur J. Bernstein.) Presented By : Megha Priyanka.
CS 540 Database Management Systems
GPFS: A Shared-Disk File System for Large Computing Clusters Frank Schmuck & Roger Haskin IBM Almaden Research Center.
KYUNG-HWA KIM HENNING SCHULZRINNE 12/09/2008 INTERNET REAL-TIME LAB, COLUMBIA UNIVERSITY DYSWIS.
Querying the Internet with PIER CS294-4 Paul Burstein 11/10/2003.
Load Rebalancing for Distributed File Systems in Clouds.
Distributed File System. Outline Basic Concepts Current project Hadoop Distributed File System Future work Reference.
The Google File System Sanjay Ghemawat, Howard Gobioff, and Shun-Tak Leung Presenter: Chao-Han Tsai (Some slides adapted from the Google’s series lectures)
Computer Science Lecture 19, page 1 CS677: Distributed OS Last Class: Fault tolerance Reliable communication –One-one communication –One-many communication.
An Introduction to GPFS
Cofax Scalability Document Version Scaling Cofax in General The scalability of Cofax is directly related to the system software, hardware and network.
Accelerating Peer-to-Peer Networks for Video Streaming
Architecture and Algorithms for an IEEE 802
Providing Secure Storage on the Internet
Building a Database on S3
Workshop in Distributed Data & Structures
The Gamma Database Machine Project
Presentation transcript:

Rim Moussa University Paris 9 Dauphine Experimental Performance Analysis of LH* RS Parity Management Workshop on Distributed Data Structures: WDAS 2002

2 1. Contribute towards improving effictiveness of the 1st prototype [M. Ljungström] : Data Bucket Split 2. Proposal of Scenarios High Availability Bucket Recovery Objectives

3 Overview 1. Why SDDS ? 2. LH* RS Data Structure 3. File Creation 4. High Availability 5. Recovery 6. Conclusion & Future Work

4 Motivation  Information Volume of 30% / year  Bottleneck of disk access and CPUs  Failures Are frequent Scalability High Performance High Availability

5 Hardware architecture Modular Architecture Modular Architecture Best cost/ Performance ratio Best cost/ Performance ratio Need Network-based Storage Systems SDDS Multicomputers

6 SDDS  Dynamic file growth Client Network Client … Data Buckets Inserts … … Coordinator … Overloaded Split Records Transfert

7 SDDS (ctnd.)  No Centralized directory access Client Network Query IAM … … … … Data Buckets Query

8 High Availability ? Distribution Nodes’ Failure Parity Calculus High storage cost Data Replication Data Replication

9 LH* RS in a few words SDDS Distribution –Hashing Function Parity Calculus –Reed Salomon Codes

10 LH* RS – File Structure       r [ e -1 … -1 ] P Insert Rank er 4 Data Buckets Key Data Field Rank Key list Parity Field

11 LH* RS : Split Scenario Splitting Data Bucket New Data Bucket e r Delete e of rank r r* e Insert e in rank r* e r* Insert e in rank r* Delete e of rank r

12 Why the use of TCP/IP ?  Flow Control  No more loss of messages even if parity sites are overloaded even if parity sites are overloaded In opposition to UDP – Ljungström thesis In opposition to UDP – Ljungström thesis  Parity Buckets coherence  Serialize Communication at PBs Critical sections Critical sections

13 Hardware Testbed  6 Pentium III, 730 MHz, 128 Mb Machines  Ethernet network: max bandwidth of 100 Mbps  1 entity: (bucket, client)/ Machine  Configuration tested: 1 Client 1 Client A group of 4 Data Buckets A group of 4 Data Buckets K Parity Buckets, k  {0, 1, 2} K Parity Buckets, k  {0, 1, 2}

14 File Creation Performances 0,40 ms 0,44 ms 0,48 ms Insert Time/ record +10% 4,9ms 6,5 ms 8 ms Ack Key 10001

15 File Size High Availabilty Degradation of the High availability of the file Solution Add a Parity Bucket/ Group

16 Parity Bucket Creation New Parity Bucket Coordinator Data Bucket’s group Insertion Your Content ? Autogenerate

17 Parity Bucket Creation Perf %79.93%65.71%57.89% Connection Time/ Total Time

18 Data Bucket Recovery UDP TCP/IP 2 Scenarios

19 Recovery Client Group g of Data Buckets Coordinator Query Group g Failure !

20 Coordinator Probe Data Buckets Parity Buckets Recovery Manager

* Recovery Manager Available Buckets r [ k 1 k 2 k 3 k 4 ] P 2 1* 2 Spare Buckets Deduce Record/ key=k 3 Deduce Record/ key=k 2 Insert Record/ key = k 2 ! Insert Record/ key = k 3 ! 3 Record/ key = k 1 ? Record/ key = k 4 ? Record/ key = r ? UDP-based recovery scenario

22 UDP-based Recovery Scenario 0.55 ms 0.65 ms 0.76 ms Just 0.1ms to compute a record/ iteration Just 0.1ms to compute a record/ iteration

* Recovery Manager Available Buckets 1* 2 Spare Buckets Deduces Records having rank  [r, slice-1] 3 Buffer of Records to Insert ! Records of rank [r, slice-1] ? Records of key [r, slice-1] ? TCP/IP-based recovery scenario

24 TCP/IP-based Recovery Scenario Communication Time >> Process Time. Communication Time >> Process Time. Slice increases implies better performance results. Slice increases implies better performance results.  b = records  file of recs  records/B

25 Discussion TCP/IP vs UDP Reliability Reliability Performance gain Performance gain Best improvements when slice = entire bucket content (31250 recs). Indeed, UDPTCP/IPGain 1 DB s 6.7 s160%

26 Conclusion  Implementation of a new split algorithm Use TCP/IP instead of UDP Use TCP/IP instead of UDP Use Critical section to manage the concurrent requests of updates at the level of Parity Buckets Use Critical section to manage the concurrent requests of updates at the level of Parity Buckets  Parity Buckets Management  Efficient Data Buckets Recovery

27 Future Work  More performance Measurements  Variation of Parity Calculus

References [LS00] [Ljungström, 2000] CERIA & U. Linkoping [Rizzo] [Luby] [XB99]

Demo of the Prototype Friday – Poster Session CERIA Lab. B017

End Thank you for your Attention