|
Paper #22
A Centralized, Tree-Based Approach to
Network Repair Service for Multicast Streaming
Media
Dan Rubenstein
University of Massachusetts at Amherst
Nicholas F. Maxemchuk & David Shur
AT&T Labs - Research
|
Abstract:
IP multicast provides best-effort delivery. Packets encounter
variable delays and may be lost because of transmission errors
and buffer overflows. Real-time multimedia streaming
services require that most packets arrive at the receivers
prior to an application deadline. Multicast quality on the
current Internet is often inadequate for these applications.
We have solved this problem by placing repair servers inside
the network. The repair servers recover missing packets
by communicating with each other, then re-multicast the re-paired
stream to nearby receivers on a new address. Multicast
reception in the constrained area is typically much better
than in the wide area Internet.
In this paper we address the problem of constructing a
repair graph. The repair graph shows which repair servers
each repair server communicates with to recover missing
messages. Our objectives when constructing this graph conflict
with each other. We want high reliability: every repair
server to recover as many missing messages as possible, as
quickly as possible. But we also want low cost: this recovery
should use as little of the network bandwidth as possible.
We present a centralized algorithm to generate re-pair
graphs. We demonstrate through simulation that these
graphs achieve a level of reliability that is almost as high as
that achieved by repair graphs specifically designed for high
reliability. At the same time, our graphs maintain a cost that
is almost as low as the cost in repair graphs designed for
low cost.
Download a copy of this paper in either:
Compressed PostScript (59K) - or -
PDF (96K) format.
- Last revised: Weds Jun 27 20:56:29 EDT 2000
