Caching Architectures for Scalable Streaming Media Delivery
<p>
Azer Bestavros (Boston University)
<p>
Abstract:
<p>
The delivery of streaming media objects presents a formidable strain on
server and network capacity due to the long duration and high bandwidth
requirements that are characteristics of streaming media workloads. In
this talk, I will start with an overview of the characteristics of
streaming media workloads, and will discuss the promise and limitations
of a number of recently proposed scalable streaming media delivery
techniques that rely on patching and periodic broadcasting. Next, I will
present two instances that illustrate how "non-traditional" caching
architectures could significantly improve the scalability and quality of
streaming media delivery. In the first, I propose the use of "partial
caching", an approach that turns edge caches into accelerators of
streaming media delivery by enabling the joint delivery of content from
caches and origin servers. A salient feature of this approach is that it
is both network-aware and stream-aware, and it takes into account the
popularity of streaming media objects, their bit-rate requirements, and
the available bandwidth between clients and servers. In the second, I
propose the use of a cache-and-relay approach to end-system multicast,
whereby a client joining a multicast session caches a streaming media
object (either partially or entirely) and, if needed, relays that stream
to neighboring peer clients, which may join the multicast session at
some later time. This cache-and-relay approach is fully distributed,
scalable, and efficient in terms of network link cost.