Dr. Naresh M. Patel, Vice President and Chief Architect at NetApp Inc. in Sunnyvale, CA, USA.
Enterprise data trends are changing rapidly in the world of hybrid multi-cloud, container-based microservices applications. Workload models for these new architectures need to change to maintain accurate predictions for how compute, network, and storage architectures will perform as the next generation technologies evolve at different rates and as workflows become more distributed and automated. The slowdown in Moore’s Law has created a renaissance in novel computer architectures. Indeed, the pendulum swings toward more disaggregation, offloads, and separation of control vs. data plane in order to continue to scale performance cost-effectively. This talk will discuss some of the key trends in enterprise data with examples from some production workloads to shed light on future modeling challenges.
Naresh Patel is a Vice President and Chief Architect at NetApp Inc. in Sunnyvale, California. He obtained his Ph.D. in Computing Science from Imperial College, London in 1989, and co-authored a book on performance modeling with Peter Harrison in 1992. After moving into industry, he has continued to pursue his interest in researching techniques for performance analysis and modeling applicable to new technologies, resulting in patents and several papers in peer-reviewed conferences and journals. In his 30+ years in Silicon Valley, he has held various leadership roles in research and development at Tandem Computers Inc., Compaq (now Hewlett Packard Enterprise), and NetApp with a focus on performance engineering of next generation systems. In his current role at NetApp, he leads architects who are driving innovation in systems, platform software and storage media. This role has provided an opportunity to anticipate future technology/market trends, to characterize production workload data, and to discover ways to help enterprises build their own next generation hybrid multi-cloud.
Prof. Amy R. Ward, University of Chicago, Booth School of Business, IL, USA.
Joint work with Prof. Amber Puha
The study of scheduling problems has a long history in the academic literature. However, many classic models used to study scheduling problems do not incorporate customer impatience. Furthermore, many of the ones that do assume that the time a customer is willing to wait for service is exponentially distributed. Our objective is to provide a methodological framework to study the impact of general distributional assumptions on scheduling decisions in the context of a many server queue (specifically, a multiclass G/GI/N+GI queue), a model commonly used to support service system design.
To do this, we specify a class of admissible control policies (rules for determining when to serve a given customer class) and formulate a fluid model. We establish a tightness result for sequences of fluid scaled state descriptors operating under admissible control policies and satisfying some mild asymptotic conditions, and show that limit points of such sequences are fluid model solutions almost surely. Then, we characterize the invariant states of the fluid model, and use that to formulate an approximate scheduling control problem. This motivates us to introduce a set of control policies, called Weighted Random Buffer Selection (WRBS), that capture the entire spectrum of invariant states in the fluid limit, and so can be used to asymptotically achieve any solution to the approximate scheduling control problem. We end with some open questions.
Amy R. Ward is the Rothman Family Professor of Operations Management at the University of Chicago Booth School of Business. She received her Ph.D. from Stanford University in 2001. She is the Stochastic Models co-Area Editor for the journal Operations Research (term began 1/2018). She recently completed her term as chair of the INFORMS Applied Probability Society (term 11/2016-11/2018). She also recently completed her term as the Service Management Special Interest Group Chair for the INFORMS Manufacturing and Service Operations Management Society (term 6/2017-6/2019).