Project Ideas for CMSC818L, Spring 2002

• Programming Networks of Embedded Systems using Smart Messages. Scale, volatility and ad-hoc configuration make networks of embedded systems impossible to program using traditional distributed computing models. Smart Messages offer an alternative solution called cooperative computing, which is essentially based on execution migration with self-routing. The goal of this project is to design a high-level programming environment for networks of embedded systems based on SM. The programs should allow the user to express the quality of results and handle routing exceptions.
• Operating System for Smart Messages. Smart Messages require a simple system support in the center of which there is a Tag Space. Memory operations, I/O and synchronization are all performed through the Tag Space. The goal of this project is to design and implement a micro OS kernel with an event-based architecture. In this OS, a Smart Message is nothing else but a mobile event handler in the Tag Space.
• Smart Message over Bluetooth. The goal of this project is to design and implement a content-based network layer over Bluetooth. This basically requires supporting scatternets. Smart Message communication will perform on top of the BT scatternets using content-based routing.
• Smart Message Security. Smart Messages supports cooperative computing through execution migration and mobile code, solutions that are obviously prone to security attacks. The current Smart Message implementation ignores security issues. In order to make it a viable support for distributed computing over networks of embedded systems, security must be addressed. The big challenge is to design a distributed security model assuming nodes with limited resources.
• Routing Schemes using Smart Messages. The key feature of the Smart Message architecture is self-routing by which we understand that routing is chosen by the application as a migration library. The application can also intervene in migration to change routing and/or destination. This topic can lead to multiple projects. The routing scheme must be original but can be inspired from existing content-based routing schemes. In all cases nodes must be assumed volatile and mobile. Of particular interest are geographical routing, global energy-aware routing, probabilistic and rendez-vous routing schemes.
• Distributed applications over Smart Messages. The goal of this project is to design and implement complex distributed applications using Smart Messages. Object-tracking using multiple intelligent camera nodes is an example of complex distributed application but other non-trivial applications are also acceptable.
• Relaxed Transport Protocols over Networks of Embedded Systems. Existing transport protocols with end-points bound to hosts and strong end-to-end semantics cannot be used for networks of embedded systems where nodes, links and routes are volatile and mobile. The goal of this project is to design and implement a new class of transport protocols with dynamic binding of end-points to hosts and dynamic splitting of connection in the network. This project does not assume Smart Message support.
• Relaxed Transport Protocols over Overlay Networks. This project mirrors the goals of the previous project but will apply to WAN rather than to networks of embedded systems. The basic idea is to decouple the transport protocol from the network layer. The protocol must support a content-based end-point naming using one of the existing P2P schemes for location and routing. In particular, this protocol should work for user-to-user communication in pervasive computing environments where users change hosts during connection.
• Split-Networking for Network Servers. This one and the following two projects are part of the Split-OS project, which aims to support server performance by splitting the operating system between the host and intelligent devices (I-NIC) or dedicated front-end nodes (connection servers). The key ingredient of this split architecture is the non-intrusive intra-server communication provided by VIA or Infiniband-like interconnects. The goal of this project is to offload the networking protocols to a front-end node using a kernel-based implementation starting from an open-source UNIX system.
• Direct Device-to-Device Communication. The goal of this project is to design and implement the kernel mechanisms to allow direct device-to-device communication bypassing the host and its memory. Applications control direct device-to-device communication but operating system must ensure reliable and protected communication between devices. The storage device must provide a DAFS server interface. Policy to decide between host-mediated vs. direct device-to-device communication must also be developed and evaluated using a real server application. This project will be performed in conjunction with the Split-Networking project.
• Cooperative Networking. This project targets to provide a shared TCP layer on top of multiple intelligent network interfaces or connection servers. Connections can be shared among multiple I-NICs and the cooperative protocol must tolerate I-NIC failures and implement load balancing. This project will be performed in conjunction with the Split-Networking project.
• Distributed DAFS. DAFS is an efficient remote file system protocol designed to take advantage of memory-to-memory communication interconnects such as VI or Infiniband. The goal of this project is to develop a distributed DAFS over a number of DAFS servers to provide a single global file system space.
• Your project idea.