Pastry: Scalable, decentralized object location and routing for 
large-scale P2P systems
---------------------------------------------------------------
A. Rowstron and P. Druschel

The paper presents Pastry, a prefix-based peer to peer protocol 
that aims at efficiently locating objects in a scalable and  
decentralized manner. Each node in Pastry has assigned a unique 
id and each message has a key associated with it. A message is 
routed to the node with an id closest to the message key. Pastry 
tries to minimize the distance that messages have to travel in 
order to arrive at their detination.

As protocol strengths, the first one is the scalability. This is 
obtained through a scheme that requires a logarithmic number of 
steps to route messages. Scalability is important is large scale 
systems where the number of nodes can be very high. 
Another thing is that, when routing messages to peers, the Pastry
protocol considers network proximity (contrary to the Chord 
protocol for example). An issue here is the proximity metric since 
this is not so trivial. They state traceroute as a possibility for 
the application to use but ICMP filters may be a problem in a 
large scale system.

Besides their simulation results, real experimental results would 
be needed to be able to fairly evaluate this protocol. In a real
system, nodes joinings and leavings are nondeterministic and may
represent a problem for the protocol performances. A comparison 
of Pastry and Chord would also be helpful. This paper and the 
Chord paper, both state the other protocol as a similar one but 
which one performs better and under which conditions?