PeerVM
From BoykinWiki
The PeerVM project is an ongoing effort at the ACIS Laboratory at University of Florida,Gainesville which brings together the recent advances in the fields of Peer-to-Peerrouting technologies,Virtualization and Grid technologies in an attempt to build a Planetary Scale Distributed, Self-Configurable, and Decentralized Virtual Infrastructure. A more formal introduction about PeerVM can be found here.
Contents |
Subsystems
Peer-to-Peer Routing
For the purpose of Peer-to-Peer routing, we use the Brunet P2P Library. We would like to emphasize here that the motivation of using Brunet as compared to other well-known and popular Peer-to-Peer systems like Chord from MIT, Pastry from Microsoft Research or the Symphony system from Stanford University lies in its ability to provide a simple and scalable solution similar to STUN in order to traverse the NATs and some simple kind of firewalls being fully integrated into the routing layer itself.
Our other motivation for using Brunet was that one of the faculty members Dr. P. Oscar Boykin who is currently associated with the PeerVM project was the originator and the chief architect of the Brunet system started at University of California, Los Angeles and hence we had sufficient knowledge at hand and an easy access to the Brunet system. However, we must point out that in order to make the system usable for our purposes, we have customized it to our requirements and have made lots of enhancements to the base system and still are in the process of making changes and adding more modules to it on a routine basis as and when need arises.
Virtual Networking Services
For providing the Virtual Networking Solutions for the PeerVM system, we currently have two flagship projects running in our lab under the titles of IPOP:IP-over-P2P and WOW respectively. For detailed discussions about them please visit the respective project pages.
Job Scheduler
Currently for the purpose of scheduling, we are using the very popular and widely used Condor system available from the University of Wisconsin-Madison. Presently, we have tried to integrate the Condor scheduler with the peer-to-peer routing layer and the IPOP and WOW virtual networking solutions and have developed a plug-and-play solution which we name as Condor Appliance for the purpose of Wide-Area Distributed Computing from personal desktops and laptops based on the freely available Player and Server products from VMware Inc. For detailed discussions and downloads please visit the Condor Appliance homepage.
However, keeping into account our vision of a Planetary Scale Virtual Infrastructure, we are looking to devise our own scheduling mechanisms which can scale upto millions of nodes. One of the novel ideas we have in this direction is to go for a "Destination-Aware Scheduling". The way this works is that the jobs are scheduled at nodes which are having a closer physical and network proximity so that the response time can be minimized. One of the early precursors in this direction is the Octant system from Cornell University which attempts to determine the physical location of Internet hosts based on their reply to the ICMP ping packets.
File System Services
We are currently exploring the design space for the file system services and are looking closely at existing Peer-to-Peer based file system solutions like Kosha from Purdue University or the Oceanstore system from University of California, Berkeley.
Security Services
The basic security features we want to integrate in our system are:
a) Allow only trusted nodes to be a part of the PeerVM system. This can be achieved by enforcing security mechanisms such as exchange of security certificates at the edges of the network when a node tries to join it.
b) Provide security to nodes from other nodes in the network with regard to their data integrity and privacy. This can be achieved by encrypting the packets at the source and decrypting them at the destination.
Resource Management
At this point we are far from needing the functionality of resource management but are observing closely the developments which are coming along.
User Interface Manager
We plan to provide a graphical User Interface Manager with a bundle of tools which will include tools for monitoring the health of the whole PeerVM network and the base PlanetLab network, tools to provide drag and drop functionality for files to users with in their Virtual Machines, networking tools using which a group of users may collaborate and create their own Virtual Private Network i.e. "PeerNets" or Virtual Private Clusters i.e. "PeerClusters". However, we must admit that we are far from this point and haven't really given much thought to this issue at the moment.
The Team
Faculty Members
Graduate Students
Abhishek Agrawal
Arijit Ganguly
Tae-Woong Choi
Amrita Nayal
David Issac Wolinsky
James Rhett Aultman
Publications
1) Abhishek Agrawal, Arijit Ganguly, P. Oscar Boykin, Renato J. Figueiredo, "Towards P2P routed IP Overlay Networks for Grid Virtual Machines", Poster (In Proceedings of the 14th IEEE International Symposium on High Performance Distributed Computing, HPDC), Research Triangle Park, NC. pdf ppt
2) Arijit Ganguly, Abhishek Agrawal, P. Oscar Boykin and Renato Figueiredo, "IP over P2P: Enabling Self-Configuring Virtual IP Networks for Grid Computing", In Proceedings of the 20th IEEE International Parallel and Distributed Processing Symposium (IPDPS), Rhodes Island, Greece. pdf
3) Arijit Ganguly, Abhishek Agrawal, P. Oscar Boykin, Renato Figueiredo "WOW: Self-Organizing Wide Area Overlay Networks of Virtual Workstations", In Proceedings of the 15th IEEE International Symposium on High Performance Distributed Computing (HPDC), Paris (to appear). pdf
Design Documents
2) Condor Appliance ReadMe File
Code Repositories & Downloads
- Brunet Code Repository
- IPOP Code Repository
- paper writing
- Condor Appliance Downloadable Images and Documents
