1. GridTorrent Framework: A High-performance Data Transfer and Data Sharing Framework for Scientific Computing

2. Outline Introduction Motivation PTCP GridTorrent Framework
Test Results
LAN Test Results
WAN Test Results
Overhead
Questions

3. Introduction Today’s computational science is data-intensive
Large Hadron Collider (LHC) experiment at CERN generates petabytes of data
Accessibility, replication and creation of the data are made very easy by
Internet
Computational Grid

4. Motivation The data is geographically distributed Users are dispersed
Collaboration environments are required
Resources should be used in efficient and effective way
CPU
Storage
Network

5. PTCP TCP has a performance problem over WANs
It was developed to solve the above problem by using striping technique

6. GridTorrent Framework
It is aimed
to provide collaboration environment for dispersed users
to make data transfer, management, and sharing easy via content manager
to use systems resources efficiently and effectively by harnessing P2P (Bittorrent) network structure

7. Collaboration & Content Manager
The Content Manager allows users to publish or share their files with selected access control rights
The Collaboration Manager permits users to build a virtual sharing environment by managing working groups or friend list
ACL enforce access control rights for a given content
Task Manager handles the users’ task list

8. GridTorrent Client It is responsible for
initiating actual data publishing
data sharing with other GTF clients
ensuring secure environment for the above activities

9. WS-Tracker WS-Tracker is a WS enabled server
It assists in the communication between peers (GridTorrent clients)
It delivers task lists which is generated by users to GridTorrent clients
It supplies ACL of each shared file to GridTorrent clients

10. Experimental Results A B C
PTCP and GridTorrent Framework tests cases were conducted both in LAN and WAN type of computer networks
Server and client machines’ specification and location table
File size is 300 MB
Name
Specification
Network Interface
Institution
Location A
Intel(R) Quad-Core Xeon(TM) 4x2.33GHz CPU with 8GB of RAM on Red Hat Enterprise Linux 4.0
Broadcom NetXtreme II BCM Base-T
Indiana University
Bloomington, IN B
Sun Fire V880 8x1.2GHz UltraSPARC III processors with 16GB of RAM on Solaris 9. It has 6x72GB 10K rpm internal HD
Gigabit Ethernet and 10/100-BaseT Ethernet C
Dual Pentium III 731MHz CPU with 512MB of RAM on GNU/Linux fc5
Florida State University
Tallahassee, FL

11. LAN Test Setup Server is located at Bloomington, IN
Client is at Indianapolis, IN
The number of parallel TCP streams between server and client has increased from 1 to 16 (PTCP)
The number of seeders increased from 1 to 16 (GTF)
Client and server configuration for PTCP
GridTorrent test case configuration for LAN
test. Regular Java sockets are used for data
transfer.

12. LAN Test Result
There is no significant improvement in bandwidth for both PTCP and GTF
Experimental data transfer ( Mbps) rate is much lower the theoretical (1000Mbps) and measured data transfer rate (857Mbps)

13. WAN Test-I Setup Server is located at Bloomington, IN
Client is at Tallahassee, FL
The number of parallel TCP streams between server and client has increased from 1 to 16 (PTCP)
The number of seeders increased from 1 to 16 (GTF)
Client and server layout for PTCP test
case
GridTorrent test case configuration for wide
area network test. Regular Java sockets are
used for data transfer.

14. WAN Test-I Result
Bandwidth usage is vastly improved in both GTF and PTCP
PTCP’s bandwidth utilization rate has risen steadily until fifteen streams
its peak value is 118 Mbps
GTF’s bandwidth utilization rate has risen steadily until thirteen streams
GridTorrent is performing better than PTCP when parallel streams number is less than five

15. WAN Test-II Setup Server is located at Bloomington, IN
Client is at Tallahassee, FL
The number of parallel TCP streams between server and client has increased from 1 to 16 (PTCP)
Besides Java socket, other data transfer protocols can be exploited in GridTorrent client
The number of seeders increased from 1 to 16 (GTF)
Four parallel TCP sockets were used between peer and seeders
Client and server layout for PTCP test
case
GridTorrent test case configuration for wide
area network test. GridTorrent client uses four
parallel TCP sockets in each connection for
every source

16. WAN Test-II Result
Using parallel TCP with Bittorrent algorithm demonstrates much better bandwidth usage than standalone GridTorrent and PTCP
The maximum attained bandwidth is around 145 Mbps which is %23 higher than PTCP’s result

17. Overhead
Both parallel TCP and GridTorrent have overhead due to nature of multiple parallel connections
PTCP’s communication channel overhead time can be compared to GridTorrent WS-Tracker client’s overhead time varying between 300 and 600 milliseconds
Another overhead of GridTorrent is that control messages exchanged between peers to ensure strictly enforced to all participating peers
The total size of overhead messages is between 148KB to 169 KB

18. Questions

19. Thanks to All.