Research Interests:
Research Interests:
Research Interests:
Research Interests:
ABSTRACT Soft errors arising from energetic particle strikes pose a significant reliability concern for computing systems, especially for those running in noisy environments. Technology scaling and aggressive leakage control mechanisms... more
ABSTRACT Soft errors arising from energetic particle strikes pose a significant reliability concern for computing systems, especially for those running in noisy environments. Technology scaling and aggressive leakage control mechanisms make the problem caused ...
Research Interests:
Research Interests:
We describe our implementation choices and experiences gained from developing a GMPLS-based control plane test-bed prototype for end-lo-end services. The prototype includes control plane components of routing and signaling protocols and... more
We describe our implementation choices and experiences gained from developing a GMPLS-based control plane test-bed prototype for end-lo-end services. The prototype includes control plane components of routing and signaling protocols and provides rapid provisioning and restoration of connections across multiple optical network domains. Our experience in implementing the multi-domain signaling showed that it was possible to simply adapt the signaling
Research Interests:
Next generation optical network technology has attracted lots of attention from both academia and industry. Large optical networks are typically partitioned into metro and core sub-networks due to the different service and technology... more
Next generation optical network technology has attracted lots of attention from both academia and industry. Large optical networks are typically partitioned into metro and core sub-networks due to the different service and technology requirements. A client optical connection may pass through multiple sub-networks, originating in a metro sub-network, traversing the core network, and accessing the destination on a different metro network. This structure is similar to current ISP networks, which may be divided into multiple routing areas for scalability. However, optical networks are different from IP networks. The difference affects the routing information that needs to be exchanged in metro/core optical networks such that IP routing protocols require significant extensions if they are to be applied in optical networks. In this paper, we focus on Metro/Core routing requirements and present a proposal to extend the current IP multi-area Open Shortest Path First (OSPF) protocol to exchange routing information over metro/core sub-network interfaces. A brief discussion of our prototype implementation is given at the end of this paper.
Research Interests:
ABSTRACT A network-to-network interface (NNI) offers multi-domain interworking via standard protocols, extending provisioning and restoration capabilities end-to-end. Our contributions are: (a) a fully functional NNI prototype (the first... more
ABSTRACT A network-to-network interface (NNI) offers multi-domain interworking via standard protocols, extending provisioning and restoration capabilities end-to-end. Our contributions are: (a) a fully functional NNI prototype (the first to our knowledge), (b) the demonstration that existing distributed protocols suffice, augmented with a few simple, novel extensions.
Research Interests:
Research Interests: Computer Science, Algorithms, Automata, Dynamics, Models, and 15 moreRelaxation, Channels, Blocking, High Speed, Interference, Geographic distribution, Parallel Algorithm, Discrete Event Simulation, Parallel Simulation, Community Networks, Event horizon, Time, Speedup, Intervals, and Continuous time
Research Interests:
Research Interests:
The problem of computing the maximum of n inputs on an asynchronous parallel computer is considered. In general, the inputs may arrive staggered in time, the number of processors available to the maximization algorithm may vary during its... more
The problem of computing the maximum of n inputs on an asynchronous parallel computer is considered. In general, the inputs may arrive staggered in time, the number of processors available to the maximization algorithm may vary during its execution, and the number of inputs, n , may be initially unknown. Two simple, efficient algorithms to compute the maximum are presented. Each algorithm may be invoked asynchronously, as new inputs and processors arrive. Performance measures that account for the response times of the invocations are introduced, and the algorithms are analyzed under these measures.
Research Interests:
Research Interests:
A ring of N cells rotates in discrete steps past N queues, moving customers from their queues of arrival to randomly chosen destinations. The model has applications in communication systems, processor interconnection networks, and... more
A ring of N cells rotates in discrete steps past N queues, moving customers from their queues of arrival to randomly chosen destinations. The model has applications in communication systems, processor interconnection networks, and flexible manufacturing. The arrivals to the queues are independent and stochastically identical. The total numbers of arrivals to the system during successive steps are independent, identically distributed random variables with mean λ and finite second and third moments. A greedy policy governs the insertion of customers on the ring: A customer waiting at the head of a queue enters the next unoccupied cell to appear at that queue. The customer then remains on the ring a random travel time d, leaves, and frees its cell for another customer A necessary condition for the system to be stable is . If no customer travels further than once around the ring is sufficient for stability. Other results assume d to be stochastically bounded by an exponentially distributed travel time with mean . Then is sufficient for stability. In the limit of large N, stable systems with fixed λ and μ have expected numbers of waiting customers per queue; then a customer's wait in a queue is usually negligible compared with his travel time. Simulations suggest that the mean number waiting in a queue may even be .
Research Interests: Mathematics, Applied Mathematics, Statistics, Flexible Manufacturing, Queueing theory, and 14 moreStability, Communication System, Asymptotics, Reliability, Ring, Stochastic models, Travel Time, Computer System, Exponential distribution, Numerical Analysis and Computational Mathematics, Discrete random variable, Asymptotic Behavior, Queue, and Cyclic
Research Interests:
Research Interests:
Modern cloud architectures rely on each server running its own networking stack to implement policies such as tunneling for virtual networks, security, and load balancing. However, these networking stacks are becoming increasingly complex... more
Modern cloud architectures rely on each server running its own networking stack to implement policies such as tunneling for virtual networks, security, and load balancing. However, these networking stacks are becoming increasingly complex as features are added and as network speeds increase. Running these stacks on CPU cores takes away processing power from VMs, increasing the cost of running cloud services, and adding latency and variability to network performance. We present Azure Accelerated Networking (AccelNet), our solution for offloading host networking to hardware, using custom Azure SmartNICs based on FPGAs. We define the goals of AccelNet, including programmability comparable to software, and performance and efficiency comparable to hardware. We show that FPGAs are the best current platform for offloading our networking stack as ASICs do not provide sufficient programmability, and embedded CPU cores do not provide scalable performance, especially on single network flows. A...
Research Interests:
Research Interests:
Research Interests:
Research Interests:
Abstract Calls that make large, persistent demands for network resources will be denied consistent service, unless the network employs adequate control mechanisms. Calls of this type include video conferences. Although overprovisioning... more
Abstract Calls that make large, persistent demands for network resources will be denied consistent service, unless the network employs adequate control mechanisms. Calls of this type include video conferences. Although overprovisioning network capacity would ...
Research Interests:
It is our great pleasure to welcome you to SIGMETRICS/Performance 2009. SIGMETRICS is the flagship conference of the ACM special interest group for the computer systems performance evaluation community. Performance is the flagship... more
It is our great pleasure to welcome you to SIGMETRICS/Performance 2009. SIGMETRICS is the flagship conference of the ACM special interest group for the computer systems performance evaluation community. Performance is the flagship conference of the IFIP working group on performance modeling and analysis. Every three years, the two conferences are held jointly, and this is the eleventh joint conference. This year, we will continue with several of the innovations introduced at last year's SIGMETRICS program. The main conference will again be a full three days, featuring 27 papers, 21 posters, and three invited talks from computer science luminaries, both academic and industrial. We will also reprise the demo competition and student thesis panel that were so well received last year. We are introducing an industrial information seminar, to provide an opportunity for students and academics to hear from representatives in industrial research about performance-related projects and their impact on deployed products and services. Supplementing the main conference are four interesting workshops, ranging from the venerable to the avant-garde. The workshop on MAthematical performance Modeling and Analysis (MAMA) continues its eleventh year as a forum for talks on early research in the more mathematical areas of computer performance analysis. The workshop on Hot Topics in Metrics (HotMetrics), which had a stellar inauguration last year, will reprise its role in helping to identify "big" and "hard" problems in performance evaluation and to develop innovative approaches to solving them. We also introduce two new workshops this year: GreenMetrics will explore how improvements to or new uses of Information and Communication Technology (ICT) can contribute towards efforts to minimize global climate change, a problem of increasing importance in modern society. The Learning for Networking workshop will investigate the use of machine learning techniques to tackle the increasingly complex architecture and control features in telecommunications and computer networks. This year also features a splendid series of tutorials, on topics ranging from social networks to data-center networks, from data-flow programming to packet-flow configuration, and from internet measurement to internet service construction.
Research Interests:
Research Interests:
Research Interests:
To create the Azure cloud, we imagined and built a massive cloud network, literally from the ground up. Azure creates and sustains a virtual data center for every tenant, under the tenant's control, meeting the tenant's goals for... more
To create the Azure cloud, we imagined and built a massive cloud network, literally from the ground up. Azure creates and sustains a virtual data center for every tenant, under the tenant's control, meeting the tenant's goals for high security, reliability, ease of use, and performance. Building all this calls for innovation across hardware and software. I'll discuss key challenges and solutions, as well as demo new functionality.
Research Interests:
ABSTRACT In this paper, we investigate the power of such broadcast in solving a paradigmatic problem in distributed computing. Imagine a network in which each node machine Ni (1-&-le;i-&-le;n) keeps a Boolean value vi in... more
ABSTRACT In this paper, we investigate the power of such broadcast in solving a paradigmatic problem in distributed computing. Imagine a network in which each node machine Ni (1-&-le;i-&-le;n) keeps a Boolean value vi in local memory. The vi 's determine a set S-&-equil;{i: vi-&-equil;1}. The non-emptiness problem on n nodes is to find some i in S, or else find that S is empty. In practice, a problem of this type arises in two ways: 1. Consensus testing: Has any node voted -&-ldquo;no-&-rdquo;, where vi-&-equil;1 means node i votes -&-ldquo;no-&-rdquo;, and vi-&-equil;O means node i votes -&-ldquo;yes-&-rdquo;? 2. Establishing a distinguished node: S specifies a set of candidates, and solving non-emptiness selects one. For example, the nodes may be bidding for a Job or a resource. In section 2, we introduce an idealistic broadcast communication scheme which abstracts certain features of the CSMA technology.
Research Interests:
In this position paper, we provide an overview of tomo-gravity and anomography, recently developed methods for fast, accurate and effective estimation of traffic matrices and associated anomalies in large operational IP networks. These... more
In this position paper, we provide an overview of tomo-gravity and anomography, recently developed methods for fast, accurate and effective estimation of traffic matrices and associated anomalies in large operational IP networks. These methods provide the fundamental inputs needed to put the engineering and operation of these networks on a sound and effective scientific footing.
Research Interests:
Research Interests:
A wheel rim clamping device is disclosed which has a plurality of pressure applying surfaces for clamping a concentric wheel rim to an axle hub assembly so as not to deform the rim and to improve the load transfer between the rim and hub.... more
A wheel rim clamping device is disclosed which has a plurality of pressure applying surfaces for clamping a concentric wheel rim to an axle hub assembly so as not to deform the rim and to improve the load transfer between the rim and hub. The clamping device, in one embodiment, includes an arcuately shaped main body portion and a plurality of alternatively offset, bent over, pressure applying surfaces along the outer longitudinal edge portion of the main body. The pressure applying surfaces grip both sides of an inwardly directed rim locator ring which extends around the interior periphery of the rim. Drive lug recesses are formed between the pressure applying surfaces to accept rim drive lugs which are spaced along the rim locator. The pressure applying surfaces and rim drive lugs lock the rim in place. The inner longitudinal edge portion of the main body includes downwardly extending and radially spaced apart depending leg portions which are mounted to the central axle hub assembly. The transverse end portions of the main body include semi-circular recesses, and the opposed semi-circular recesses on the end portions of a pair of side by side clamping devices form a mounting opening for an adjusting bolt. The adjusting bolt provides for both adjustability between the clamping devices and a more uniform clamping force against the rim by forcing the ends of adjacent clamping devices outwardly against the rim. A plurality of the clamping devices are mounted around the rim locator end-to-end thereby forming a substantially continuous annular mounting ring assembly between the hub and concentric wheel rim. In a second preferred embodiment, each of the pressure applying surfaces is arcuately shaped with a generally rectangular perimeter to provide more gripping area against the complementary arcuate sides of the rim locator.
Research Interests:
Thesis (Ph. D.)--University of Washington, 1983. Vita. Includes bibliographical references (leaves [123]-125).
Research Interests:
Research Interests:
ABSTRACT The electromechanical characteristics of a novel micro torsion-mirror actuator and the optical properties about the light scattering and reflection of its micro-mirror under the influence of mirror surface roughness are... more
ABSTRACT The electromechanical characteristics of a novel micro torsion-mirror actuator and the optical properties about the light scattering and reflection of its micro-mirror under the influence of mirror surface roughness are investigated experimentally and theoretically. The electrostatic yielding voltages for driving the metal-coated polysilicon or single crystal silicon micro-mirrors suspended by elastic torsion beams to tilt from 0 degree(s) to 90 degree(s) spontaneously are in the range of 270~290 V and the minimum holding voltages for keeping the tilting angle of the mirrors to be in 90 degree(s) are found 55 V or so when the thickness of the torsion-beams is about 1 mm. The lifetime and the estimated shortest actuating time of the micro torsion-mirror actuators can reach 108 at least and less than 2 ms, respectively. The maximum value of the surface roughness of the micro-mirrors is 69.443 nm and the distribution of the surface roughness is smooth, which is acceptable for wavelength division multiplexing applications basically. In general, the novel torsion-mirror actuators can be applied as optical switches for optical networking.
Research Interests:
ABSTRACT Soft errors arising from energetic particle strikes pose a significant reliability concern for computing systems, especially for those running in noisy environments. Technology scaling and aggressive leakage control mechanisms... more
ABSTRACT Soft errors arising from energetic particle strikes pose a significant reliability concern for computing systems, especially for those running in noisy environments. Technology scaling and aggressive leakage control mechanisms make the problem caused ...
Research Interests:
RECAL is a new recurrence relation for calculating the partition function and various queue length moments for closed, product form networks. In this paper we discuss a number of the issues involved in the software implementation of RECAL... more
RECAL is a new recurrence relation for calculating the partition function and various queue length moments for closed, product form networks. In this paper we discuss a number of the issues involved in the software implementation of RECAL on both sequential computers and parallel, shared memory computers. After a brief description of RECAL, we describe software implementing RECAL on a sequential computer. In particular, we discuss the problems involved in indexing and data storage. Next we describe code implementing RECAL on a parallel, shared memory computer. Special attention is given to designing a special buffer for temporary data storage and several other important features of the parallel code. Finally, we touch on software for serial and parallel implementations of a tree algorithm for RECAL.
Research Interests:
Research Interests:
We present SGuard, a new fault-tolerance technique for distributed stream processing engines (SPEs) running in clusters of commodity servers. SGuard is less disruptive to normal stream processing and leaves more resources available for... more
We present SGuard, a new fault-tolerance technique for distributed stream processing engines (SPEs) running in clusters of commodity servers. SGuard is less disruptive to normal stream processing and leaves more resources available for normal stream processing than previous proposals. Like several previous schemes, SGuard is based on rollback recovery [18]: it checkpoints the state of stream processing nodes periodically and restarts failed nodes from their most recent checkpoints. In contrast to previous proposals, however, SGuard performs checkpoints asynchronously: i.e. , operators continue processing streams during the checkpoint thus reducing the potential disruption due to the checkpointing activity. Additionally, SGuard saves the checkpointed state into a new type of distributed and replicated file system (DFS) such as GFS [22] or HDFS [9], leaving more memory resources available for normal stream processing. To manage resource contention due to simultaneous checkpoints by di...
Research Interests:
Research Interests:
Research Interests:
Gauss-Seidel is a general method for solving a system of equations (possibly nonlinear). It makes repeated sweeps through the variables; within a sweep as each new estimate for a variable is computed, the current estimate for that... more
Gauss-Seidel is a general method for solving a system of equations (possibly nonlinear). It makes repeated sweeps through the variables; within a sweep as each new estimate for a variable is computed, the current estimate for that variable is replaced with the new estimate immediately, instead of on completion of the sweep. The idea is to use new data as soon as it is computed. Gauss- Seidel is often efficient for computing the invariant measure of a Markov chain (especially if the transition matrix is sparse), and for computing the value function in optimal control problems. In many applications the computation can be significantly improved by appropriately ordering the variables within each sweep. A simple heuristic is presented here for computing an ordering that quickens convergence. In parallel processing, several variables must be computed simultaneously, which appears to work against Gauss-Seidel. Simple asynchronous parallel Gauss-Seidel methods are presented here. Experimen...
Research Interests:
Methods for rapid, high-throughput sequencing of nucleic acids (DNA) are evolving at a breakneck pace to meet the future needs in the field of individualized genetic medicine.
Research Interests:
A problem related to the decentralized control of a multiple access channel is considered: Suppose k stations from an ensemble of n simultaneously transmit to a multiple access channel that provides the feedback 0, 1, or 2+, denoting k =... more
A problem related to the decentralized control of a multiple access channel is considered: Suppose k stations from an ensemble of n simultaneously transmit to a multiple access channel that provides the feedback 0, 1, or 2+, denoting k = 0, k = 1, or k ≥ 2, respectively. If k = 1, then the transmission succeeds. But if k ≥ 2, as a result of the conflict, none of the transmissions succeed. An algorithm to resolve a conflict determines how to schedule retransmissions so that each of the conflicting stations eventually transmits singly to the channel. In this paper, a general model of deterministic algorithms to resolve conflicts is introduced, and it is established that, for all k and n (2 ≤ k ≤ n ), Ω( k (log n )/(log k )) time must elapse in the worst case before all k transmissions succeed.
Research Interests:
Pipelining is now a standard technique for increasing the speed of computers, particularly for floating-point arithmetic. Single-chip, pipelined floating-point functional units are available as off the shelf components. Addressing... more
Pipelining is now a standard technique for increasing the speed of computers, particularly for floating-point arithmetic. Single-chip, pipelined floating-point functional units are available as off the shelf components. Addressing arithmetic can be done concurrently with floating-...