apan2004-kath
Download
Report
Transcript apan2004-kath
A P2P Distributed
Computing Platform:
Korea@Home
2004.1.28(Wed)
C.Y. ‘Connor’ Park
[email protected]
KISTI Supercomputing Center
Overview
Korea@Home (KatH) : A Software
Platform for Distributed Computing
using Volunteer Computer Resources
Features
Hybrid Peer-to-Peer architecture
Secure PKI-based Security and Digital
Signature
Web Services with SOAP messaging
2
Motivation
To take advantage of and stimulate
‘Broadband Network Infrastructure in
Korea’
To meet the growing needs of
computing power for applications
Funded by ‘Ministry of Information and
Communication (MIC)’
3
Broadband Network Infrastructure
in Korea
Over 10M Broadband network (xDSL, Cable)
households (over 70% penetration)
Over 58.4% of PCs with Pentium3+
Own PCs
Not own PCs
100%
80%
23.1
21.4
100%
Internet-unable
17.7
10.7
82.3
89.3
2001.12
2002.12
80%
60%
40%
Internet-enable
– 2002.11
60%
76.9
78.6
40%
20%
20%
0%
0%
2001.12
2002.12
4
Similar Projects
Science
Internet
SETI@Home,
climateprediction.net
Life Science
Grub, CycleTrader
Financial
MoneyBee
Folding@Home,
FightAIDS@Home
Cryptography
Distributed.net,
RSAttack576
Games
ChessBrain
Miscellaneous
DALiWorld
Mathematics
Mersenne Prime Search,
Pi(x) Project
from http://www.aspenleaf.com/distributed/index.html
5
Project History
2002. 4.
MIC launched the pilot project ‘Korea@Home’
Application : Virtual Screening for New Drug Design
2002. 8.
First pilot production and test
Client/Server Architecture Model
2002. 12.
Second pilot production and test
2003. 4.
Regularize the project “P2P Internet-based Distributed
Computing Platform Development”
2003. 9.
Announcement of Hybrid P2P-style Distributed Computing
Platform
6
Architecture
Agent
Agent
Cooperation
Agent
Cooperation
Distribution
Distribution
Cooperation
Distribution
Agent
Internet
Distribution
Subscription
Job Request
Resource
Management
Large-scale
Application
7
Job Distribution
Agent
Agent
Agent
Authentication
Internet
Working Specification
Request
• Is there any job for
me to work?
• Who has the job
Agent
data?
Getting data, working,
and returning result to
server
KatH Server
8
2002 Pilot Projects
Application : Virtual Screening for New Drug
Design
Client-Server Architecture
An Agent
Receives about 400KB-size file
Executes during 20 minutes per working unit
Returns about 190KB-size file
Total number of working job units : 50,000
9
Virtual Screening
Target Protein
Chemical DB
Share
Analysis
Disease Protein
Clinical Demo New Drug Mass-produce
Ligand DB
Virtual High-Throughput Screening
+
+
Client 1
+
+
Client 2
Server
+
Calculating Active Site PMI
Calculating Ligand PMI
+
Looking for
Drug Candidate
Client 3
Internet
Structure Analysis
10
Experimental Results
1st Test
1,217 installed agents
2002.8.22-9.2
Max 53.5 Gflops / Avg. 31.78 Gflops
2nd Test
6,472 installed agents
2003.2.6-2.14
Max 644 Gflops / Avg. 470.6 Gflops
※ World Top Rank 124 - 629.7 Gflops:
IBM xSeries Cluster Xeon 2.8 GHz - Myrinet/ 192
(from http://www.top500.org, 2003.6)
11
Intention of Participation
Korea@Home Understanding
Don’t Know (33.3%)
Intention of Providing Resources
Providing with
Don’t (2.9%)
Compensation (7.2%)
Well-Understanding
(44.9%)
Keep up
Providing (59.5%)
Providing later (30.4%)
Understanding
roughly (21.8%)
• 67.7% of Netizen already know what is Distributed Computing and
Korea@Home.
• 89.9% of Netizen have intention of providing their PC resources.
12
2003 Progresses
P2P data distribution
New Applications
Virtual Screening for New Drug Design
Global Risk Management for financial
investment
3D image rendering
Supplementary plug-in
Remote backup service
13
Futurework
Make P2P agents more intelligent and
cooperative with each other.
A set of APIs for application developers
Fault-tolerance : checkpointing &
rollback recovery
IPv6 version
Open-Source (?)
Windows-only Unix version
14
Thanks a lot! ^^
[email protected], http://www.koreaathome.org
15
Design Goals
Application Developers (Job Requester)
Convenience
They should be able to use easily.
Flexibility
The platform is general enough to be used by
various applications.
Correctness
The platform should return correct results.
16
Design Goals
PC agents (Resource Providers)
Transparency
PC users are not disturbed in their own works.
Reliability
It’s reliable not to make any security and privacy
problem.
Reward
There is any kind of compensation, even just in
mind.
17
Design Goals
Platform
Scalability
Performance should be improved according to
the growth of agents.
Adaptability
It should put up with the dynamic features of
Internet.
18
Implementation Choices
Make server as simple as possible,
and Make PC agents more intelligent
Web Services
SOAP messaging over https
Request-Response Model
It is Agents’ duty to make most decisions.
PKI-based certification and Digital
Signature for files
19
Functional Classification
Resource Management
Authentication, Certification
Simple accounting
Scheduling and Job distribution
Job distribution with hybrid P2P style
Result gathering
Client Management
Client authentication
Application preparation to work
Digital signing
Result returning to client
20
Networking Issues
Network Security
DoS, Spoofing, Sniffing, …
Variable Connection
Dial-up, cable, xDSL, T1/E1, …
IPv6, WiFi, Bluetooth, …
Network Unobtrusiveness
Background/shared communication
Transparency (Firewall/NAT)
21