VMware is pleased to announce that Rashmi Vinayak, Assistant Professor of Computer Science at Carnegie Mellon University (CMU), is the 2021 recipient of the VMware Systems Research Award. Professor Vinayak is recognized for her research enhancing computer system reliability and efficiency through a synergistic co-design of coding-theoretic algorithms, machine-learning models, and systems. In recognition of her originality, impact, and future potential, the VMware University Research Fund will gift CMU $125,000 to support Professor Vinayak’s research.
Much of Professor Vinayak’s work has involved coding theory, which is the study and application of mathematical constructs for transforming data from one form to another in communications and storage. Codes can be used for myriad purposes — resilience, efficiency, confidentiality — and are key building blocks for modern computing. Especially in this current era of big data and data analytics, codes are foundational to the systems that underpin modern industry and society. Even small improvements in our ability to store, recall, analyze, and transmit data can have significant impact for individuals, VMware’s enterprise customers, and the overall environmental sustainability of our industry.
Professor Vinayak is known both as a theoretician and as a systems researcher. Says Professor Muriel Medard of the Massachusetts Institute of Technology (MIT), “Rashmi has brought coding- and information-theory-inspired techniques into a successful systems suite of solutions with great originality and intellectual power.”
Professor Aditya Akella of University of Texas observes, “Rashmi perfectly straddles the theory-systems boundary, making fundamental contributions in both areas and producing impactful real-world systems that are grounded in foundational new theory.”
This broad expertise was first developed in Professor Vinayak’s Ph.D. work and concerned IO-efficient storage codes. The “piggyback” scheme she developed with her collaborators involves systematic augmentations of existing codes (e.g., storage efficiency) while adding new and useful properties (e.g., repair efficiency). The practical value of these codes has been recognized and implemented in important systems. “Rashmi’s papers on how to use coding convinced the open-source community of the viability of applying these results in the Hadoop File System (HDFS),” says Dhruba Borthakur, former HDFS project lead and a continuing reviewer of contributions to HDFS.
Although she began with theory, Professor Vinayak is known for working closely to understand and develop systems that will be impactful in real-world settings. An important reason she stayed to do a postdoc was because she wanted to find domains in which she could apply coding theory.
One example includes collaboration with Google researcher Arif Merchant. Arif and his colleagues have been collaborating with Professor Vinayak to create storage codes that account for the failure rate of individual disks. These codes can save as much as 20% of the disks, which aligns cost savings and sustainability benefits when applied to some of the world’s largest storage clusters. Considering the rapid growth and importance of big data, these results can confer significant societal benefit. “She has a sharp, incisive mind and is quick to pick up the critical parts of the problem and come up with interesting directions,” says Arif.
Another example is Professor Vinayak’s collaboration with Twitter, which resulted in a large-scale analysis of hundreds of in-memory key-value cache clusters. This work is gaining recognition because it explains both what is happening and why. The “Segcache” project proposes a new storage layout for in-memory caches and has the potential to reduce Twitter’s memory footprint up to 60%. It is slated to go into production soon and stands as testimony to the sustainability impact of her work.
Professor Vinayak is currently looking at the relationship between machine learning (ML) and coding theory. ML is amongst the most important new workloads of our times. And because modern applications are increasingly distributed, it has become clear that the intersection of coding, ML, and distributed computation will surface both new problems and new opportunities. Professor Vinayak’s team has been working at this intersection. One of her recent breakthroughs has applied coding and ML to limit the impact of failures and slowdowns in machine-learning systems. The work, “Learning-based Coded Computation,” supports general nonlinear functions, as opposed to supporting only linear and polynomial functions. The results may be used to demonstrate improvements in neural-network inference systems.
The future presents rich opportunities for coding-related systems research to make an enormous difference. “Both data and compute needs are growing exponentially,” says VMware Fellow Pratap Subrahmanyam. “Applications with global scope need to be operated reliably and efficiently as we strive for sustainability. Professor Vinayak’s work holds a great deal of promise toward useful progress on that journey.” VMware Fellow Christos Karamanolis adds, “Professor Vinayak has shown the ability to do leading work in numerous communities, from theory-oriented to systems-oriented. With the combination of fundamental theory, a deep interest in practical applications, and systems research thinking, it’s not surprising to see her making an impact at diverse top-tier venues like OSDI, FAST, SOSP, SIGCOMM, ISIT, ICML, and AISTATS.”
The VMware Systems Research Award celebrates early-career faculty within the first five years of a first tenure-track appointment. This year, the selection committee was chaired by Princeton Professor Jennifer Rexford and included Professor Anastasia Ailamaki of EPFL (known in English as Swiss Federal Institute of Technology, Lausanne), University of Washington Professor Tom Anderson, Georgia Institute of Technology Professor (and Dean of the College of Computing) Charles Isbell, UCB’s Professor Sylvia Ratnasamy, VMware’s Christos Karamanolis and Pratap Subrahmanyam, as well as myself. We see the award as a way to support and give back to the academic research community, which plays a crucial role in exploring new technology.