|
Decentralized Resourse Management & Job Scheduling for Scalable & Reliable Cloud Computing Platform
|
[ Overview ]
- Current Grid/Cluster Computing Platform is based on traditional server-client approcach for resource management and job scheduling.
We adopt a current popular peer-to-peer technology for resource managment and job scheduling for scalabla and reliable system.
These decentralized and distributed approach can be used to build a large-scale cloud computing platform becasue pure p2p approach does not have problems of single points of failures or performance bottlenecks.
Distributed Hash Table (DHT),one of famous peer-to-peer techbnology,
is widely used in real ysystems such as file sharing, distributed file system, and contents delivery networks.
We use Content-addressable Networks (CAN) among various DHTs for our resrouce management and job scheduling purpose.
|
[ Results ]
- Efficient Resource Management Scheme
- Effective Load Balancing Scheme
- Performance competitive to a Centralized Scheduler
|
[ Publication ]
- Michael Marsh, Jik-Soo Kim, Beomseok Nam, Jaehwan Lee, San Ratanasanya, Bobby Bhattacharjee, Peter Keleher, Derek Richardson, Dennis Wellnitz and Alan Sussman, "Matchmaking and Implementation Issues for a P2P Desktop Grid", 2008 National Science Foundation Next Generation Software Workshop (NSFNGS '08) in conjunction with IPDPS 2008, Miami, FL, April 2008
- Jaehwan Lee, Pete Keleher and Alan Sussman, "Decentralized Multi-attribute Range Search for Resource Discovery and Load Balancing", The 23rd International ACM Symposium on High-Performance Parallel and Distributed Computing (HPDC'12), submitted
|
Fig.1. Decentralized Resource Management (P2P + Grid Computing)
|
|
|
|