Project Suggestions

Note: Many of the projects below are large enough to support multiple groups.
• An Operating System Kernel for Swarm Computing. Swarm computing is a new computing paradigm enabled by the development of mass-producible micro-electro-mechanical devices (MEMs) with limited computing and (wireless) communication capabilities. Thousands of such devices, for example, can be sprayed around a volcano crater, submerged to the ocean floor, or injected into a human bloodstream. They would typically perform measurements and coordinate actions (such as emitting certain chemicals or even performing cooperative micro-surgery). These devices would typically run a suite of coordination protocols which control their communication, relay information to and from outside observers, interact with the physical environment, and provide suitable high-level abstractions to programmers and operators of the swarm. The collection of these protocols may define the next generation operating system for swarm computing. In this project, we shall lay the main architecture of such operating systems and implement a chosen subset of Swarm communication and coordination protocols that belong to it.

• Priority-aware UNIX Sockets for High Performance Servers. UNIX is the most popular server platform today. Servers typically listen to a well-known port for client connection requests. These requests are queued in FIFO order until they are accepted by the server. In reality, clients may have different priorities and should be handled accordingly by the operating system. This project has two parts. First, UNIX socket implementation will be modified to support priority queuing of incoming client requests to a high-performance server. Second, a new operating system interface will be designed for applications (i.e., servers) to define how priority values should be assigned to requests coming off the network interface. The modified operating system will be tested by running an Apache Web server on top and demonstrating that high priority clients suffer less queuing delays as a result of the new extensions.

• Group Communication Protocols for Weak Consistency. Group communication protocols ensure consistency of global state in a distributed system by providing communication primitives that implement virtual synchrony. In many cases, virtual synchrony is an overkill. For example, a set of cooperative web caches need only to be loosely synchronized. Loose synchrony allows a bounded deviation of the local member state from the global state. A member can purchase a ``de-synchrony ticket'' from the communication protocol. The ticket allows the member to alter its state locally by a certain maximum amount specified in the ticket without causing communication messages to other group members. Once the maximum state deviation is exceeded, the communication subsystem produces a broadcast message re-synchronizing the member with the global state. Hence, the ``inconsistency'' of each member remains bounded at all times by a user-defined parameter. Successful protocols for implementing parameterized weak consistency may be the key to applications such as replicated e-commerce, reservation, and trading servers. The next generation of such applications would allow online users to interact with and alter the state of local replicas without going through centralized bottleneck components. This may be the enabler of active Web caching technology; a technology that allows on demand caching (i.e., replication) not only of static files but also of scripts that generate dynamic Web content.

• Operating System Support for Computing under Uncertainty. Unmanned, autonomous, possibly intelligent, robotic vehicles, actuators, and probes are increasingly needed for operating in high risk environments such as exploring planetary surfaces, deep underwater exploration, emergency operations in disaster areas, and military defense applications. Such vehicles will be computer controlled. They will perform their missions in uncertain, unfriendly, highly dynamic, and perhaps poorly modeled environments. They will have to execute multiple (computational) tasks and make time critical decisions under uncertainty. In the absence of sufficient information and time, an operating system would have to allocate resources to these tasks in such a way that maximizes the probability of mission success. A key architectural feature in such an operating system would be the use of sensory feedback from the environment for adapting resource allocation. In an analogy with human psychology, the next generation feedback-based operating system will implement the ``reflex'' actions of the autonomous system. Application programs (possibly involving AI techniques) will implement higher level ``cognitive'' actions that are slower in response but are more ``intelligent''. In this project we shall develop the architectural vision for the new operating system and implement a simple prototype to be used for mission-critical computing under uncertainty.