Bay Area Research Wireless Access Network (BARWAN)
Technical Approach Overview
IP and TCP/UDP provide the basic network and transport layers upon
which overlay and wireline networks will be integrated. Nevertheless,
new interfaces are needed that reveal more about the underlying
performance capabilities of the networks to applications. Effective
management of overlay networks requires protocol extensions that
reach down to the network layer. To build the extendable and scalable
mobile information systems we envision, able to deliver high quality
connectivity to mobile applications over the wide-area, we are addressing
the following challenges:
- Seamless Integration of Overlay Networks:
No general network management
architecture exists for effectively integrating multiple overlay networks.
Mobile applications roaming across overlays requires network intelligence
to determine that the mobile has moved from acceptable coverage in one
network to better coverage in another. But a global network management
algorithm is stilled needed to control handoffs across overlays based on
current mobile connectivity. Link quality is only one metric that determines
handover; priority of access, applications needs, and relative cost are
equally important. Since overlays may not cooperate with one another to
render such decisions, we will demonstrate mobile assisted handoff in
which the mobile host must be an active participant in handoff processing.
- Support Services for Mobile Applications:
Handover across overlays will change an application's network bandwidth
and latency. We will design a new applications interface to the network
management layers to allow them to initiate handovers, to determine
changes in their current network capabilities, and to gracefully adapt
their communications demands. We will better integrate mobile applications
and scalable wireline processing and storage capabilities through an agent
processing architecture that exploits data type specific transmissions to
manage the communications demands over dynamically varying wireless links.
- Managing Mobile Connections to Support Latency-Sensitive Applications:
Handoffs must be executed with lower latency than is now possible if (near)
real-time multimedia applications are to be well supported. One strategy
moves the routing and resource allocations to local subnets. For example,
roaming authentications can be cached locally to avoid repeated remote,
latency-intensive transactions. We will develop algorithms that exploit
information about the location of mobile devices, the geographical
adjacency of cells, and the likely routes devices might take, to improve
handoff processing. End-to-end strategies like Mobile IP provide routing,
but fall far short for latency-sensitive connection-oriented services.
More hierarchical approaches, which localize information collection to
the region or the subnet containing the users, are more scalable.
- Load Balancing for Scalable Mobile Processing:
Repositioning within future wireless networks will be a common event.
Traffic patterns will not be uniform, with high correlations between
mobile host location, their repositionings, and their requests for
service. We are developing network management architectures that build
on decentralized algorithms to allocate network and processing resources
on demand, avoiding the static and centralized schemes of the past.
Furthermore, overlay networks provide an opportunity to share bandwidth
and processing across networks; current network load is one reason to
initiate internetwork handoff.
Randy H. Katz, ed., randy@cs.Berkeley.edu; Last edited: 21 APR 95