A Highly Available Cluster Storage System Using ... - Semantic Scholar

A Highly Available Cluster Storage System Using Scavenging High Availability and Performance Computing Workshop (HAPCW'04), held with 2004 Los Alamos Computer Science Institute Symposium (LACSI’2004)

Xubin (Ben) He, Li Ou

Stephen L. Scott, Christian Engelmann

{hexb, lou21}@tntech.edu

{scottsl, engelmannc}@ornl.gov

Storage Technology & Architecture Research(STAR) Lab Electrical and Computer Engineering Department

Computer Science and Mathematics Division

Outline

Introduction Highly Available Metadata Management Metadata Management using Bloom Filters Conclusions and Future Work

A Highly Available Cluster Storage System Using Scavenging, HAPCW'2004

2

NOW Handout Page 1 CS258 S99

1

Motivations • Data intensive scientific applications Æ high performance computing. • Distributed storage systems: high performance storae, wide area mass storage, cluster storage. • Excellent performance, parallel support ÅÆadministration costs, central points of failure and control

A Highly Available Cluster Storage System Using Scavenging, HAPCW'2004

3

Existing work: • HPSS • Peer-to-Peer Storage • Scavenging [Vazhkudai]

Grid Data Access

Grid Storage Management

Scavenger Registration

Scavenged Storage Morsels

A Highly Available Cluster Storage System Using Scavenging, HAPCW'2004

4

NOW Handout Page 2 CS258 S99

2

9

Introduction Highly Available Metadata Management Metadata Management using Bloom Filters Conclusions and Future Work

A Highly Available Cluster Storage System Using Scavenging, HAPCW'2004

5

User Data vs. Metadata Metadata manager

Storage server

Client

A Highly Available Cluster Storage System Using Scavenging, HAPCW'2004

6

NOW Handout Page 3 CS258 S99

3

Distributed Metadata Management Client Access

Cluster Storage Management

Scavenge Manager Unified Interface

Metadata Management Layer

Scavenged Storage Morsels

High Availability: Scheme #1 (Pure P2P)

Peer-To-Peer Network

Metadata

Manager

NOW Handout Page 4 CS258 S99

4

High Availability: Scheme #2 (active/hot-standby)

Peer-To-Peer Network

Hot-Standby Manager

Active Manager

Metadata

High Availability: Scheme #3(Partitioning) Master/Load Redirector

Network

Metadata

Manager

NOW Handout Page 5 CS258 S99

5

High Availability: Scheme #4(leader/helper)

Network

Metadata

Helper

Leader

Group Communications • Peer-to-Peer distributed control • Reliable broadcast, Atomic Broadcast • Atomic transactions guarantee metadata integrity

A Highly Available Cluster Storage System Using Scavenging, HAPCW'2004

12

NOW Handout Page 6 CS258 S99

6

9

Introduction Highly Available Metadata Management Metadata Management using Bloom Filters Conclusions and Future Work

A Highly Available Cluster Storage System Using Scavenging, HAPCW'2004

13

Bloom Filter • A Bloom filter is a fast and efficient method for representing a set A={a1,a2,…,an} of n elements to support membership queries. Element a

1

h1(a)=P1

h2(a)=P2 h3(a)=P3

1 1

m bits

h4(a)=P4

1

A Highly Available Cluster Storage System Using Scavenging, HAPCW'2004

14

NOW Handout Page 7 CS258 S99

7

Metadata Management using Bloom Filters File queries Scavenge Manager Unified Interface

Hashing and LRU Cache

hit

Update

miss hit

Bloom Filter Array

miss Multicast the queries

15

Introduction Highly Available Metadata Management Metadata Management using Bloom Filters 9 Conclusions and Future Work

A Highly Available Cluster Storage System Using Scavenging, HAPCW'2004

16

NOW Handout Page 8 CS258 S99

8

Conclusions •

Investigate Availability issues Scavenged Storage Systems, propose 4 solutions for maintaining multiple metadata managers: – – – –

•

P2P Active/hot-standby Partitioning Leader/Helper

Speed up the metadata searching: –

Bloom Filters

A Highly Available Cluster Storage System Using Scavenging, HAPCW'2004

17

Future Work

• Comparing the proposed 4 schemes • Scalability • Metadata Cache

A Highly Available Cluster Storage System Using Scavenging, HAPCW'2004

18

NOW Handout Page 9 CS258 S99

9

Acknowledgements • Research Office and Center for Manufacturing Research, Tennessee Technological University • Ralph E. Powe Junior Faculty Enhancement Award by Oak Ridge Associated Universities (ORAU). • Mathematics, Information and Computational Sciences Office, Office of Advanced Scientific Computing Research, Office of Science, U. S. Department of Energy.

A Highly Available Cluster Storage System Using Scavenging, HAPCW'2004

19

Questions and Comments?

A Highly Available Cluster Storage System Using Scavenging, HAPCW'2004

20

NOW Handout Page 10 CS258 S99

10

A Highly Available Cluster Storage System Using Scavenging High Availability and Performance Computing Workshop (HAPCW'04), held with 2004 Los Alamos Computer Science Institute Symposium (LACSI’2004)

Xubin (Ben) He, Li Ou

Stephen Scott, Christian Engelmann

{hexb, lou21}@tntech.edu

{scottsl, engelmannc}@ornl.gov

Storage Technology & Architecture Research(STAR) Lab Electrical and Computer Engineering Department

Computer Science and Mathematics Division

NOW Handout Page 11 CS258 S99

11