Report from Internet2/SURA P2P conference
by Andy Oram
Last week I visited a surprisingly chilly Phoenix, Arizona to deliver
at the workshop
Collaborative Computing in Higher Education: Peer-to-Peer and
I no longer think that P2P requires its own distinctive research, but
I still find that people trying to meet the needs of P2P
infrastructure and applications are doing some of the most interesting
work in computing. Here are some key points I heard at the
conference. A longer article with more general points will be posted later this week.
The value of a peer-to-peer mindset is shown in a modest project
Your Own Internet Distribution,
Its goal is to produce an efficient multicast system for networks that
don't support IP multicast.
According to Lindell, a typical way someone might add users to a
multicast would be to build a mesh in which all users are fully
connected—for instance, ten (4+3+2+1) connections for five users.
Then the application would perform some sophisticated routing analysis
and come up with a hierarchical tree that represents the best routes
between all users. The problem is that the application would have to
rebuild the mesh and the tree every time a user comes or goes, and
that creates burdensome overhead in a real-life situation.
A P2P mindset takes account of transient users and designs a solution
that is less elegant in theory but more workable in practice. No mesh
is created, but each user that connects is added to the best point in
the tree. If one node becomes overwhelmed with traffic, it kicks one
or more children off the tree and they rejoin it elsewhere.
Bill St. Arnaud,
as well as other presenters, emphasized that there are plenty of
legitimate applications, especially peer-to-peer applications, that
require high bandwidth. So bandwidth problems won't go away, even if
all the file-sharing sites are shut down.
St. Arnaud, who helps Canadian universities develop fiber networks, is
in a highly unusual position because his colleagues have plenty of
bandwidth and are looking for applications to fill it. His
showed some promising educational projects.
Most university administrators, in contrast, really need to rein in
file sharing. It's driving normal research- and class-oriented traffic
off of their networks. The most promising technologies for "traffic
shaping" don't involve blocking particular ports (because file sharing
programs can work around that) or hard limits on usage (because a lot
of network users have big bandwidth needs for good reasons related to
university business). Instead, sophisticated traffic shaping watches
each users' bandwidth use and gradually cuts down on the bandwidth
allocated to them as usage increases.
Two projects illustrate the validity of a peer-to-peer approach in
ways that are particularly easy to see:
presented by CEO
is a commercial venture about which I wrote a
last year. Porivo installs its software on participating end-user
sites to test the responsiveness of web sites under absolutely
a research project presented by
lets people find authorities on particular subjects while maintaining
the privacy of both the authorities and the requesters. Each expert
stores a profile on his or her personal computer and the profile is
queried when anyone sends a request to the network for help.
described a video distribution system (an application pursued by lots
of P2P researchers) that depends on storing chunks of videos on
users's systems, and
that pushes new articles of interest through a chain of users.