Computer Systems

Computer Systems provides the foundation upon which all other software applications rely. The goal of systems research is to develop the key abstractions and services that enable software to be efficiently and portably run on hardware. Areas of interest to the systems group include operating systems, computer networks, parallel and distributed computation, and computer security.
The systems group tackles problems from both theoretical and experimental approaches. To support our experimental work, the group maintains several laboratories in the Computer Science department and in UMIACS. The Laboratory for Parallel and Distributed systems includes a collection of parallel computers and clusters to support systems research. Current equipment includes a 24 processor SPARC SMP, 8 processor IBM Power 4 system, and a 128 processor Myrinet-connected Linux cluster. The Distributed Systems Software Laboratory contains flexible networking environment to allow students to configure networking switches to allow for experimental research. In addition, the laboratory includes about 20 machines to support experiments.
The history of the systems group at Maryland dates back more than 30 years. One early member of the group, Yohan Chu, wrote the book "Computer Organization" which was the first major book on the subject. This book was used extensively at many universities and colleges. David Mills was an early researcher in computer networks. He set up an ARPANet IMP (predecessor of the current Internet) in his basement using a PDP 11/45 (an early mini-computer). At the time, this was the only full ARPANet node not located at a University or a Government facility. In the late 1970's, Chuck Rieger and Mark Weiser built ZMOB, an early parallel computer based on commodity microprocessors. The system consisted of 128 Z-80 processors.
Students and PostDocs from the systems group have gone on to faculty and industry positions around the world. Recent graduate students in faculty positions include: Gagan Agrawal (Ohio State University), Suman Banerjee (University of Wisconsin), Ugur Cetintemel (Brown University), Ibrahim Matta (Boston University), Bongki Moon (University of Arizona), Ron Larsen (Dean of College of Information Science, University of Pittsburg), and Sang Son (University of Virginia). Many of our former students have gone on to careers at major research labs including AT&T Labs (Vijay Gopalakrishnan , Seungjoon Lee), Google (Ruggero Morselli), and IBM T.J.Watson (Henrique Andrade, I-Hsin Chung, Andrzej Kochut, Kyung Ryu). The group also has a rich history of PostDoc researchers who have gone onto successful careers. For Example Anurag Acharya, and Guy Edjlali are now at Google.
Current faculty in the systems group include:
Bill Arbaugh's speciality is information security and privacy. In the past, information systems were large, expensive, barely inter-connected, and non-mobile. Today, information systems are small, lightweight, highly connected via wireless technology and mobile. Tomorrow, they will be even smaller and more mobile. The constant evolution in the design and operation of information systems presents new and increasingly complex challenges for computer security. One technology he developed is Copilot, a security and management monitor that is capable of detecting potential intruders in a high assurance manner and ensures that the intruders activities does not compromise the main system. He is also the CEO of a start-up company, Komoku, Inc.
Ashok Agrawala, an AAAS Fellow, researches the basic nature of information and it's implications to the design and implementation of computer systems. He developed an Information Dynamics Framework, which distinguishes between information and its representation, recognizing that computers only deal with the representations. In 2004, he won the University of Maryland's "Invention of the Year" for Horus Technology. Horus, a novel location determination technology developed with Moustafa Amin Youssef, uses unique algorithms to efficiently process the signal information which is used to determine position.
Bobby Bhattacharjee's research interests are in the design and implementation of wide-area networking, distributed systems, and security protocols. His current focus is on the design of decentralized secure systems for multi-party applications and large scale data distribution, especially in the context of peer-to-peer and overlay systems. His group has build systems that demonstrate protocols for scalable media streaming, randomized resilience, anonymous communication, bulk data delivery, multicast rekeying, distributed directory service, unstructured lookup, secure lookup, and predicate-based search. In current work, he is working on applying techniques from game theory and mechanism design to problems in wireless networking and peer-to-peer systems.
Jeff Hollingsworth's research is in the areas of tools for high performance computing, program instrumentation, and programmer productivity. His work on high performance computing includes tools to measure the performance of parallel programs and Active Harmony a system to allow runtime automatic tuning of parameters and algorithms for applications and middleware. His work on program instrumentation includes the Dyninst tool suite. Dyninst provides multi-platform instrumentation of binaries via both offline binary editing and online program modification. His work on programmer productivity seeks to understand how parallel programmers spend their time, and how tools and programming environments can be improved to increase productivity of parallel programmers.
Pete Keleher's research is primarily in the field of distributed systems. His work spans shared memory protocols, adaptive grid schedulers, peer-to-peer systems (including distributed searching and ranking algorithms), and gossip-based consensus algorithms. Currently, much of his work is in the context of storage systems, specifically a wide-area file system called MoteFS. MoteFS is a novel system structured around the concepts of lightweight snapshots and principal-free capabilities. The combination allows extremely efficient and fine-grained control over access to the data storage.
Udaya Shankar's research interests are in the design and analysis of distributed systems and network protocols, in both correctness and performance aspects. His correctness work deals with compositional methods for specification, verification, and testing of concurrent (including distributed) systems, focusing on realistic problems, especially at the transport and routing layers. He's currently working on a "programmer-friendly" compositional framework called SeSF that can be used for system design as well be incorporated in existing concurrent programming languages (Java, C#). His performance work deals with analytical, simulation and experimental evaluation of queuing models of networking systems, under both steady-state and transient conditions. He's currently working on timestep stochastic simulation (TSS), developing a method to compute sample paths of general queuing networks with state-dependent delayed feedback (e.g., TCP/IP/WLAN networks) with the accuracy of packet-level simulation but at a cost several orders cheaper.
Neil Spring's current research focuses on the design of network protocols that are self diagnostic: that embed features that expose and explain faults directly to users and administrators. Neil uses implementation as a primary means of evaluating his research ideas: he has constructed software to measure ISP network topologies efficiently and accurately, software to support arbitrary but safe network measurement, and software that couples instrumentation probing with data transfer in the nearly-ubiquitous TCP. To provide useful conclusions about network structure and behavior requires the aggregation of various sources of information: routing protocol information from BGP, information embedded in host names in DNS, information returned by routers in packet identifiers, source addresses, and record route entries, etc. Each source of information can, by virtue of varied implementation and configuration decisions, provide false data; a current challenge is building the reasoning logic to find the best explanations for measured Internet data.
Alan Sussman's main research area is in software tools for high performance parallel and distributed computing, which is now widely known as Grid computing. Within that broad area, one major research interest relates to interoperability of parallel (and sequential) programs, in particular how that can be applied to complex coupled physical simulations. That interest ties in closely to related interests in software component technologies, in particular how they can be applied to high-end supercomputing applications. Another major research interest involves various types of runtime and compiler support for high performance data intensive applications, and their relation to high performance database systems. A more recent research area is in applying peer-to-peer computing techniques to high-end computing problems, particularly focusing on utilizing desktop computers effectively to perform large-scale computations. All the parts of my research program are strongly motivated by high-end applications, with a current emphasis on space science, astronomy and earth science applications, motivated by collaborations with scientists working in those areas.
The Systems group receives support from the Department of Defense, Department of Energy, NASA, and the National Science Foundation. Additional support is provided by industrial partners including DoCoMo, Fujitsu, IBM, Microsoft, Samsung, and Sun Microsystems.