GCC 2003 Presentation
Sang Min Park
, Jai-Hoon Kim, and Young-Bae Ko
Ajou University
South Korea
Ajou University, South Korea
GCC 2003 Presentation
Introduction to Data Grid
Optimizations in Data Grid
Novel Replication Strategy based on Internet Hierarchy
Simulation
Simulation Results
Conclusions
2 Ajou University, South Korea
3
Data Grid Motivations
Petabyte scale data production
Distributed data storage to store parts of data
Distributed computing resources which process the data
Two Most Important Approaches for Data Grid
Secure, reliable, and efficient data transport protocol
(ex. GridFTP)
Replication (ex. Replica catalog)
Replication
Large size files are partially replicated among sites
Reduce data access time
Application Scheduling, Dynamic replication issues are emerging
Ajou University, South Korea
GCC 2003 Presentation
User
Typical Job Execution Scenario
Job Broker
Internet and
Data Grid
Grid site
CE SE
Job Distribution
Grid site
CE SE
CE
Grid site
SE
CE
Grid site
SE
Data Fetch and
Replication
Grid site
CE
Grid site
SE
Grid site
CE SE
Initial replication
Data from
Experiments (HEP,
Bioinformatics, etc)
Master Site
Large-size
Storage
Ajou University, South Korea 4
GCC 2003 Presentation
GCC 2003 Presentation
Reducing the Overall Job Execution Time
Scheduling Optimization
Deciding where to allocate the job
Considering location of replicas and computational capabilities of sites
Short-term Optimization
Deciding from where to fetch replicas
Considering available network bandwidth between sites
Long-term Optimization (Dynamic Replication Strategy)
Shortage of storage in a site
Deciding which file should be remaining as a replica
Better to replicate popular files because of its future usage
5 Ajou University, South Korea
GCC 2003 Presentation
Replica Optimization based on Site-level Locality
Replicate the file that is predicted to be used in future from the perspective of a site
Try to reduce the number of fetch
Delete Oldest, Delete LRU Method
Economic Strategy from European Data Grid
Developing OptorSim –Data Grid Optimization Simulator
Using Auction Protocol to trigger Long-term Optimization
Site-level Locality based on File access patterns
6 Ajou University, South Korea
GCC 2003 Presentation
The Limitations of the site-level optimization
A Site certainly have limitations of their storage size , which means that the rate of data request locality is also limited
There should be predictable file access patterns , but we do not know if there will be.
7 Ajou University, South Korea
GCC 2003 Presentation
Network-level Locality
A site is not the only possible source of locality
Another source of locality : Network-level locality
8
Slow Replica
Transmission
Fast Replica
Transmission
Network Region
(e.g., a country)
If the replica is located in a close site, not long delay would be taken to fetch this replica
Ajou University, South Korea
GCC 2003 Presentation
Bandwidth Hierarchy
Network Region Network Region
Network
Region
Narrow bandwidth
Contending network traffic
Grid Site
Router
Data moving path between sites within region
Data moving path between sites across regions
Ajou University, South Korea 9
GCC 2003 Presentation
Maximizing Network-level locality
1. Avoiding Replica Duplication in a region
No space here!
We should remove some file
A
Receiving New
Replica
X a Site
Delete this one!
Replica X is duplicated here!
X a Site
A Region
2. Considering popularity of file request at the region-level
10 Ajou University, South Korea
OptorSim
Data Grid Dynamic Replication Simulation tool
Developed as part of European Data Grid Project
Implemented in Java
GCC 2003 Presentation
Implemented Our own Region-based Optimizer in OptorSim
11 Ajou University, South Korea
Simulation Environment
Region A Region B
GCC 2003 Presentation
Region C Region D
Grid Site
Router
Master Site
Ajou University, South Korea
Intra-region link
Inter-region link
Link to master
12
General configuration of parameters
Parameters
Number of jobs
Number of job types
Number of file accessed per job
Size of single file
Total size of files
Values
1000
50
15
1 GB
750 GB
Bandwidth and Storage Size
Parameters
Intra-region bandwidth
Inter-region bandwidth
Master-router bandwidth
Storage space at site
Values
1000 Mbps
1000 Mbps
2000 Mbps
50 GB
Ajou University, South Korea 13
GCC 2003 Presentation
14
60000
50000
40000
30000
20000
10000
0
33174.9
47962.2
50070.9
BHR Delete LRU
Dynamic Replication Strategy
Delete Oldest
Total Job times of three strategies
Ajou University, South Korea
GCC 2003 Presentation
GCC 2003 Presentation
60000
50000
40000
30000
20000
10000
0
BHR
Delete LRU
Delete Oldest
600 800 1000 1200 1400 1600 1800 2000
Bandwidth of Inter-region link
70000
60000
50000
40000
30000
20000
10000
0
30
BHR
Delete LRU
Delete Oldest
50 70 90
Size of SE
110 130
Total job time with varying bandwidth and storage size
15 Ajou University, South Korea
GCC 2003 Presentation
The existing dynamic replication strategies are based only on site-level locality of file request
BHR strategy is based on the network-locality
BHR shows quite good performance when hierarchy of bandwidth clearly appears, and size of storage at a site is small
We extend current site-level replica optimization study to more scalable way
16 Ajou University, South Korea
GCC 2003 Presentation
William H. Bell, David G. Cameron, Luigi Capozza, A. Paul Millar, Kurt Stockinger, and
Floriano Zini.: Simulation of Dynamic Grid Replication Strategies in OptorSim. In Proc. of the 3rd Int'l. IEEE Workshop on Grid Computing (Grid'2002), Baltimore, USA, November
2002. Springer Verlag, Lecture Notes in Computer Science.
William H. Bell, David G. Cameron, Ruben Carvajal-Schiaffino, A. Paul Millar, Kurt
Stockinger, and Floriano Zini.: Evaluation of an Economy-Based File Replication Strategy for a Data Grid. In International Workshop on Agent based Cluster and Grid Computing at
CCGrid 2003, Tokyo, Japan, May 2003. IEEE Computer Society Press.
Mark Carman, Floriano Zini, Luciano Serafini, and Kurt Stockinger.: Towards an Economy-
Based Optimisation of File Access and Replication on a Data Grid. In International
Workshop on Agent based Cluster and Grid Computing at International Symposium on
Cluster Computing and the Grid (CCGrid'2002), Berlin, Germany, May 2002. IEEE
Computer Society Press.
Ann Chervenak, Ian Foster, Carl Kesselman, Charles Salisbury and Steven Tuecke.: The
Data Grid: Towards an Architecture for the Distributed Management and Analysis of Large
Scientific Datasets. Journal of Network and Computer Applications , 23:187-200, 2001.
EU Data Grid Project: http://www.eu-datagrid.org
17 Ajou University, South Korea
GCC 2003 Presentation
I. Foster, C. Kesselman and S. Tuecke.: The Anatomy of the Grid: Enabling Scalable Virtual
Organizations. International J. Supercomputer Applications, 15(3), 2001.
Wolfgang Hoschek, Javier Jaen-Martinez, Asad Samar, Heinz Stockinger and Kurt
Stockinger.: Data Management in an International Data Grid Project. 1st IEEE/ACM
International Workshop on Grid Computing (Grid'2000), Bangalore, India, Dec 2000.
OptorSim – A Replica Optimizer Simulation: http://edg-wp2.web.cern.ch/edgwp2/optimization/optorsim.html
Sang-Min Park and Jai-Hoon Kim.: Chameleon: A Resource Scheduler in a Data Grid
Environment. 2003 IEEE/ACM International Symposium on Cluster Computing and the Grid
(CCGRID'2003), Tokyo, Japan, May 2003. IEEE Computer Society Press.
Kavitha Ranganathan and Ian Foster.: Design and Evaluation of Dynamic Replication
Strategies for a High Performance Data Grid. International Conference on Computing in
High Energy and Nuclear Physics, Beijing, September 2001.
Kavitha Ranganathan and Ian Foster.: Identifying Dynamic Replication Strategies for a High
Performance Data Grid. International Workshop on Grid Computing, Denver, November
2001.
18 Ajou University, South Korea