Our work is focused on network security. We are currently working on two projects at the moment. The first, SymTrack, is an upstream firewall to limit malicious traffic based on packet symmetry. We capture the implicit signalling of reply packets from the remote host as the willingness of the remote host to accept more data from the sender.
The second project explores a new approach to intrusion detection that includes application semantics in the construction of valid workflows through an enterprise network. The system collects abstract application events from various nodes (servers) in the network, determines causal relationships (workflow) between the collected events, and raises alarms for resource accesses that do not conform to a global access control policy.  

MayaJala is a framework to implement, deploy, and use multiparty communication paradigms such as multicast (many-to-many, one-to-many), anycast (one-to-any), and gather (many-to-one). Use of such communication paradigms simplifies the development of applications with complex communication patterns. However, implementation of these communication paradigms over the Internet is a complex task and application programmers should be shielded from that complexity. MayaJala allows implementation and deployment of communication paradigms transparent to the distributed applications that use them.

We are currently working on the development of a distributed data storage system that is completely decentralized and that can operate efficiently in highly dynamic environments such as peer-to-peer systems. As part of this project, we have designed an automated redundancy maintenance engine based on durability instead of availability. The system does not offer any guarantee with respect to data being immediately available but instead provides guarantees that data will never be lost and will eventually be retrievable; an acceptable tradeoff for systems such as archival and backup storage. This approach allows to store data at a much cheaper cost in terms of network bandwidth. Preliminary results were published at the 2004 SIGOPS European Workshop and we are currently working on a full paper to present a deeper analysis and latest results.

We are experimenting with bridging disconnection-prone and intermittenly-connected networks using a layer of globally-unique persistent names on top of IP addresses. These names make proxies first-class entities in the network. This allows us to explicitly involve them in a connection without application-specific protocols.

We are trying to develop a novel approach to identify the source of a network attack and take appropriate action in such a way that can frustrate the attacker. This requires developing a sense of trust in the network for the attacked host and establishing proof of the attack so the action can be justified. If network resources can join hands, then they can be more effective against intrusions - that's the idea behind the approach.

We are working on the Routing and MAC layers in mobile ad hoc networks. Our work focuses on design and implementation of a backbone routing scheme, DCDS, to deal with uniformly distributed transient traffic. Like other backbone routing algorithms, it uses proactive components to group nodes into clusters and set up the backbone, and then deploys the backbone to reactively discover route from the source to the end. However, the reactive routing function has been carefully designed to make the backbone more robust, especially when dealing with the short-term connections. These changes enable the backbone to fulfill local recovery more efficient and propagate route changes more aggressively, while keeping the routing information accurate.

We are working on extending the Qstream framework to behave gracefully under high processor loads. This is most useful on devices with slow processors (such as portable devices) or on modern systems trying to display several videos simultaneously (such as video surveillance or video conferencing with several people at once).

In this project we are developing a common model and integration infrastructure to support application interoperability with ubiquitous computing environments across administrative and network domains. The Ubicomp Common Model (UCM) is based on a comprehensive survey and analysis of ubicomp middleware systems. This analysis led to a taxonomy of common abstractions as the basis for our entity-centric ubicomp environment model design. To evaluate this model, the Ubicomp Integration Framework (UIF) uses the UCM to expose ubicomp environments to applications using web services. Semantic web technologies are integrated into this meta-middleware platform to describe the structure of the framework in terms of entities (people, places and things) entity relationships, associated services, context and events. Current work includes additional prototype applications, and integration with other representative ubicomp systems.

We are extending an adaptive multimedia streaming system, QStream. We hope to implement a failure recovery protocol to deal with failed nodes in a multicast session. The goal is to recover from failure seamlessly without interruption to the video.

Latency tolerant parallel applications could utilize additional resources available even in high latency/high loss scenarios such as on the Internet. The transport protocol SCTP has been showed to perform quite resiliently in such scenarios, so in this work, we have incorporated SCTP and its features into the LAM implementation of MPI middleware. Using SCTP provides a closer semantic matching to MPI than TCP and can often result in increased performance for parallel applications.