(Georgia Institute of Technology, 2004)
Srinivasan, Sridhar; Zegura, Ellen W.
Many applications can benefit from the use of multicast to distribute
content efficiently. Due to the limited deployment of network-layer multicast,
several application-layer multicast schemes have been proposed. In these schemes,
the nodes in the multicast tree are end systems which are typically connected to
the network by a single access link. Transmissions to the children of a node in the
multicast tree have to share this single uplink, a factor largely ignored by previous
work.In this work, we examine the effect of access link scheduling on the latency
of content delivery in a multicast tree. Specifically, we examine the general case
where multiple packets (comprising a block of data) are sent to each child in turn.
We provide an analytical relation to compute the latency at a node in the multicast
tree and show the relationship to the packet size and block size used to transfer
data.We propose heuristics for tree construction which take link serialization into
account. We evaluate this effect using simulations and experiments on the Planet-
Lab network and show that using larger block sizes to transfer data can reduce
the average finish time of the nodes in the multicast tree at the expense of slightly
increased variance.
(Georgia Institute of Technology, 2003)
Merugu, Shashidhar; Srinivasan, Sridhar; Zegura, Ellen W.
Our work examines the role of overlay topology on
the performance of unstructured peer-to-peer systems. We focus
on two metrics of performance: (a) search protocol performance, a
local gain perceived directly by a user of the system and (b) utilization
of the network, a global property that is of interest to network
service providers. We present a class of overlay topologies based
on distance between a node and its neighbors. We show, by simulation,
that a particular topology instance of this class where every
node has many close neighbors and few random neighbors exhibits
better properties than other examined instances. In this overlay
topology, the chances of locating files are high and the nodes where
these files are found are, on average, close to the query source. This
improvement in search protocol performance is achieved while decreasing
the traffic load on the links in the underlying network.
We propose a simple greedy algorithm to construct such topologies
where each node operates independently and in a decentralized
manner to select its neighbors.