review of "Long tail"

This paper discusses the Long Tail feature of multimedia objects from a
business point of view. It is not a technique paper, but we can learn some
interesting points and better understand the workload in p2p systems.

Market has limited prediction of consumers’ interests. The popularity is
not the only factor of a movie or book making money or not. A large number
of unpopular objects also have a chance to be interested by consumers. The
reason of unpopular may be the availability is not high enough for
consumers to get it. The paper proposes three rules of how to make benefit
from unpopular objects in the “long tail”: keep availability for all
objects; split big objects into pieces; and lead consumers’ interests to
recommendations and alternations.

We can borrow ideas from this paper and re-study the workload traces in
p2p systems. Users in p2p systems are similar to consumers in
entertainment market. The workload in p2p systems also has a “long tail”.
That is, a large number of objects are requested a small number of times.
>From this paper, we can say that the behavior of p2p clients can be
redirected from popular objects to the long tail. For example, when a peer
processes a request of popular objects, it can recommend objects with
similar topic and style.

It is also important to provide availability for all objects in the p2p
system so that to improve the rate of success request. The p2p systems use
redundancy to improve that availability. That is, multiple replicas are
maintained for an object in the p2p network. The replication schema could
be different to objects with different popularity. For example, the most
popular objects are naturally distributed, no special replication are
needed. The most unpopular objects own very litter requests. More
replications for them are not necessary. The objects in the middle which
are part of the long tail need to be replicated. At least one replica is
maintained in an on-line peer, so that clients can download the objects.

It is also interesting to split big files into pieces in p2p. Sometimes, a
client only interests in a small part of a big file, but he/she has to
download the whole object in order to get the part he/she is interested
in. Bandwidth can be saved if the object is split and only the small part
is downloaded by the client. It also can improve the availability of an
object. Even though no peer holds the whole objects, a client can get it
from multiple peers who hold part of the objects. The idea is already
applied in BitTolerant and some p2p systems.
Received on Fri Nov 25 2005 - 21:56:13 EST