From: Chuck Reeves (creeves_at_windows.microsoft.com)
Date: Wed Mar 03 2004 - 13:32:28 PST
The paper, "Storage management and caching in PAST, a large-scale,
persistent peer-to-peer storage utility" was written by Peter Druschel
and Antony Rowstron in 2001. As it's name implies, it presents the
design of a peer-to-peer file system built around the Pastry routing and
location protocol. Files stored in this system are considered to be
immutable, due to the naming scheme used. While delete operations are
not supported, the storage space associated with a published file may be
reclaimed. Using the Pastry routing facilities, requests to retrieve
files in the system are routed to a node hosting the document in less
than O(log n) number of hops. Publication into PAST is controlled
through the use of certificate-based cryptography. The smartcards also
appear to act as a kind of currency for storage, where the balance of
storage available is retained securely on the card. Key expiration and
performance are interesting problems noted by the text.
The description of the Pastry algorithm was difficult to follow in the
abbreviated form presented in section 2.2. I found the paper,
"Pastry.Scalable.Decentralized.Object.Location.and.Routing.for.Large.Sca
le.Peer.to.Peer.Systems" by the same authors to be more effective in
understanding the protocol. Pastry determines storage patterns by
matching hash generated node and content identifiers. The log(n) search
behavior is further optimized through the use of proximity metric that
weights links in each nodes routing tables.
Other observations.
1. Different from Gnutella and Napster as it does address search
behavior.
2. Completely decentralized.
3. Strong guarantees of persistence.
4. Better guarantees of network locality than Chord through the use
of a scalar proximity metric.
5. File or object level store (not pages like Chord).
6. Presents an approach for quotas and security based on public key
cryptography and smart-cards.
7.
"Pastry.Scalable.Decentralized.Object.Location.and.Routing.for.Large.Sca
le.Peer.to.Peer.Systems" is better to read for implementation details.
Chuck Reeves, creeves_at_microsoft.com
Microsoft | Windows | Directory Services
This archive was generated by hypermail 2.1.6 : Wed Mar 03 2004 - 13:32:37 PST