From: ahemavathy (ahemavathy_at_hotmail.com)
Date: Wed Mar 03 2004 - 16:38:41 PST
PAST is an Internet Based, peer to peer global storage utility. The interesting feature of it is, complete decentralization. To achieve high availability, load balancing and storage utilization management techniques like replica diversion, file diversion and caching has been implemented. Every file stored in PAST are given a unique fileID at the time of creation. A client must know the fileID to retrieve the file. A posted file can be reclaimed but not deleted i.e it is not guaranteed that the file is no longer available. The files are stored on 'k' no. of nodes in the internet. Each node has a unique ID and it has a table comprising : a leaf set, routing table and neighborhood set.
The basic idea is each file is stored in k nodes whose ids are closest to the fileID. If there is less space in a node then that node can alternatively choose one of nodes from its leaf set to store the file and maintains a pointer. This is replica diversion. When a file insert operation fails, the file is given a new file ID and thus diverted to different sets of nodes. This is file diversion. Caching is another means to reduce routing time and network traffic. Local copies are cached in nodes and these are removed whenever a new file is added to the node to reclaim the space. These techniques improved the storage utilization without much overhead.
The paper was interesting and easy to read. I couldn't follow the Pastry routing but could just get a rough idea of it.
This archive was generated by hypermail 2.1.6 : Wed Mar 03 2004 - 16:38:42 PST