Building High Performance Cloud Infrastructures for the Era of Big-Data and Multicore

Dr. Xiaoning Ding
Computer Science Department, NJIT


Abstract

The cloud has become a backbone computing platform for big-data processing and many other applications. The essence of cloud computing is to share resources for lower cost and to scale applications for higher performance. However, virtualization, as the cornerstone technology for enabling and managing resource sharing in the cloud, has not been built to effectively support applications to achieve scalable performance. Focusing on multicore processors and multicore architecture, the talk will introduce fundamental design flaws in existing virtualization technology, which seriously limit application scalability and performance in the cloud. Specifically, with existing CPU virtualization designs, multicore applications cannot effectively coordinate their computation on virtual CPUs, and cannot concentrate CPU resources to reduce bottleneck. With existing memory virtualization designs, multicore applications cannot effectively localize and accelerate memory accesses by leveraging the architectural and performance features of memory devices. Existing storage virtualization is tightly coupled with CPU virtualization, leading to low storage utilization. The talk will also introduce our solutions to the above problems. *Streaming recording link: https://njit.webex.com/njit/ldr.php?RCID=1c6273260ec3ac77f16cddcf13a5975f *Download recording link: https://njit.webex.com/njit/lsr.php?RCID=f35bc76c63d5d550f5f7b193232f1ab0