A5060850653- Cloud Computing and Resource Management
- Distributed systems and fault tolerance
- Distributed and Parallel Computing Systems
- Parallel Computing and Optimization Techniques
- Peer-to-Peer Network Technologies
- Stochastic Gradient Optimization Techniques
- Optimization and Search Problems
- Age of Information Optimization
- Auction Theory and Applications
- Advanced Data Storage Technologies
- Blockchain Technology Applications and Security
- Advanced Database Systems and Queries
- Advanced Queuing Theory Analysis
- Data Management and Algorithms
- Artificial Intelligence in Games
- Data Stream Mining Techniques
- Game Theory and Voting Systems
- Caching and Content Delivery
- Scientific Computing and Data Management
- Game Theory and Applications
University of Waterloo
2022-2023
Duke University
2014-2019
Sandia National Laboratories
2019
With the democratization of cloud and datacenter computing, users increasingly share large hardware platforms. In this setting, architects encounter two challenges: sharing fairly multiple resources. Drawing on economic game-theory, we rethink fairness in computer architecture. A fair allocation must provide incentives (SI), envy-freeness (EF), Pareto efficiency (PE).
Computational sprinting is a class of mechanisms that boost performance but dissipate additional power. We describe architecture in which many, independent chip multiprocessors share power supply and sprints are constrained by the chips' thermal limits rack's limits. Moreover, we present computational game, multi-agent perspective on managing sprints. Strategic agents decide whether to sprint based application phases system conditions. The game produces an equilibrium improves task...
We study a dynamic social choice problem in which an alternative is chosen at each round according to the reported valuations of set agents. In interests obtaining solution that both efficient and fair, we aim maximize long-term Nash welfare, product all agents' utilities. present analyze two greedy algorithms for this problem, including classic Proportional Fair (PF) algorithm. several versions how they relate, provide axiomatization PF. Finally, evaluate on data gathered from computer...
Task colocation improves datacenter utilization but introduces resource contention for shared hardware. In this setting, a particular challenge is balancing performance and fairness. We present Cooper, game-theoretic framework task that provides fairness while preserving performance. Cooper predicts users' preferences finds stable matches between them. Its colocations satisfy encourage strategic users to participate inshared systems. Given Cooper's colocations, penalties are strongly...
The trend in datacenter computing is toward large, shared hardware platforms, which poses two challenges to architects: sharing fairly and multiple resources. Drawing on economic game theory, the authors rethink fairness computer architecture propose Resource Elasticity Fairness find fair allocations that ensure incentives, envy-freeness, Pareto efficiency, strategy proofness large systems.
Sharing computational resources amortizes cost and improves utilization efficiency. When agents pool their resources, each becomes entitled to a portion of the shared pool. Static allocations in round can guarantee entitlements are strategy-proof, but efficiency suffers because do not reflect variations agents' demands for across rounds. Dynamic allocation mechanisms assign multiple rounds while guaranteeing entitlements. Designing dynamic is challenging, however, when strategic benefit by...
With the democratization of cloud and datacenter computing, users increasingly share large hardware platforms. In this setting, architects encounter two challenges: sharing fairly multiple resources. Drawing on economic game-theory, we rethink fairness in computer architecture. A fair allocation must provide incentives (SI), envy-freeness (EF), Pareto efficiency (PE). We show that Cobb-Douglas utility functions are well suited to modeling user preferences for cache capacity memory bandwidth....
Computational sprinting is a class of mechanisms that boost performance but dissipate additional power. We describe architecture in which many, independent chip multiprocessors share power supply and sprints are constrained by the chips’ thermal limits rack’s limits. Moreover, we present computational game, multi-agent perspective on managing sprints. Strategic agents decide whether to sprint based application phases system conditions. The game produces an equilibrium improves task...
We present a processor allocation framework that uses Amdahl's Law to model parallel performance and market mechanism allocate cores. First, we propose the Amdahl utility function demonstrate its accuracy when modeling from core allocations. Second, design based on optimizes users' bids for processors workload parallelizability. The entitlements guarantee fairness yet outperforms existing proportional share algorithms.
With the democratization of cloud and datacenter computing, users increasingly share large hardware platforms. In this setting, architects encounter two challenges: sharing fairly multiple resources. Drawing on economic game-theory, we rethink fairness in computer architecture. A fair allocation must provide incentives (SI), envy-freeness (EF), Pareto efficiency (PE). We show that Cobb-Douglas utility functions are well suited to modeling user preferences for cache capacity memory bandwidth....
Computational sprinting is a class of mechanisms that boost performance but dissipate additional power. We describe architecture in which many, independent chip multiprocessors share power supply and sprints are constrained by the chips' thermal limits rack's limits. Moreover, we present computational game, multi-agent perspective on managing sprints. Strategic agents decide whether to sprint based application phases system conditions. The game produces an equilibrium improves task...
Computational sprinting is a class of mechanisms that boost performance but dissipate additional power. We describe architecture in which many, independent chip multiprocessors share power supply and sprints are constrained by the chips' thermal limits rack's limits. Moreover, we present computational game, multi-agent perspective on managing sprints. Strategic agents decide whether to sprint based application phases system conditions. The game produces an equilibrium improves task...
Sharing computational resources amortizes cost and improves utilization efficiency. When agents pool their together, each becomes entitled to a portion of the shared pool. Static allocations in round can guarantee entitlements are strategy-proof, but efficiency suffers because do not reflect variations agents' demands for across rounds. Dynamic allocation mechanisms assign multiple rounds while guaranteeing entitlements. Designing dynamic is challenging, however, when strategic benefit by...
Ensuring fairness in a system with scarce, preferred resources requires time sharing. We consider heterogeneous few “big” and many “small” processors. allocate processors using novel token mechanism, which frames the allocation problem as repeated game. At each round, users request big spend if their is granted. analyze game optimize users’ strategies to produce an equilibrium. In equilibrium, allocations balance performance fairness. Our mechanism outperforms classical, fair mechanisms by...
Computational sprinting is a class of mechanisms that boost performance but dissipate additional power. We describe architecture in which many, independent chip multiprocessors share power supply and sprints are constrained by the chips' thermal limits rack's limits. Moreover, we present computational game, multi-agent perspective on managing sprints. Strategic agents decide whether to sprint based application phases system conditions. The game produces an equilibrium improves task...
We consider the problem of balancing load among servers in dense racks for microsecond-scale workloads. To balance such settings, tens millions scheduling decisions have to be made per second. Achieving this throughput while providing latency is extremely challenging. address challenge, we design a fully decentralized load-balancing framework, which allows collectively system. model interactions as cooperative stochastic game. find game's parametric Nash equilibrium, and implement algorithm...
We consider the problem of balancing load among servers in dense racks for microsecond-scale workloads. To balance such settings tens millions scheduling decisions have to be made per second. Achieving this throughput while providing latency and high availability is extremely challenging. address challenge, we design a fully decentralized load-balancing framework. In framework, collectively system. model interactions as cooperative stochastic game. find game's parametric Nash equilibrium,...
With palpable effects of grid computing in different areas science, it is essential to address the need for mainstream access this rapidly growing technology. Effectiveness and accessibility a cluster greatly impacted by its ease use user interface. A robust job submission control mechanism facilitates optimum usage cluster. OSCAR portal aims answer such paper describes features benefits Web Portal. The provides central point both users administrators. underlying idea integrate...
Founded in 2017, Algorand is one of the world's first carbon-negative, public blockchains inspired by proof stake. uses a Byzantine agreement protocol to add new blocks blockchain. The can tolerate malicious users as long supermajority stake controlled non-malicious users. achieves about 100x more throughput compared Bitcoin and be easily scaled millions nodes. Despite its impressive features, lacks reward-distribution scheme that effectively incentivize nodes participate protocol. In this...
We consider the problem of balancing load among servers in dense racks for microsecond-scale workloads. To balance such settings, tens millions scheduling decisions have to be made per second. Achieving this throughput while providing latency is extremely challenging. address challenge, we design a fully decentralized load-balancing framework, which allows collectively system. model interactions as cooperative stochastic game. find game's parametric Nash equilibrium, and implement algorithm...
Sharing computational resources amortizes cost and improves utilization efficiency. When agents pool their resources, each becomes entitled to a portion of the shared pool. Static allocations in round can guarantee entitlements are strategy-proof, but efficiency suffers because do not reflect variations agents' demands for across rounds. Dynamic allocation mechanisms assign multiple rounds while guaranteeing entitlements. Designing dynamic is challenging, however, when strategic benefit by...
Computational sprinting is a class of mechanisms that boost performance but dissipate additional power. We describe architecture in which many, independent chip multiprocessors share power supply and sprints are constrained by the chips' thermal limits rack's limits. Moreover, we present computational game, multi-agent perspective on managing sprints. Strategic agents decide whether to sprint based application phases system conditions. The game produces an equilibrium improves task...