TIME: 3:30 PM

LOCATION: GMCS 314

SPEAKER: Jianshu Liu, San Diego State University, Computer Science

ABSTRACT: Scalability is an important feature of cloud computing, allowing web service providers to dynamically add or remove computing resources for their applications according to their needs. Efficiently scaling a cloud application to match workload is critical because the application must meet stringent Service- Level Objectives (SLOs), such as bounded response time, while handling naturally bursty workloads. Although various autoscaling approaches have been proposed to dynamically adjust hardware resources, they barely discuss the reallocation of soft resources (e.g., threads or database connections) after system scaling. To mitigate performance degradation caused by imbalances between hardware and soft resources, we focus on designing and optimizing soft resource adaptation to complement existing hardware-only scaling approaches.

In this talk, we show that soft resource allocation plays an important role in system performance. Various factors (e.g., hardware resource scaling, system state drift, and varying external workload types) may shift the optimal soft resource allocation at runtime. We present online models that can quickly identify the near-optimal soft resource allocation and frameworks that coordinate runtime soft resource adaptation with hardware-only scaling solutions in cloud systems. Our solutions can effectively mitigate response time fluctuations compared to state-of-the-art hardware-only scaling frameworks.

BIO: Jianshu Liu is an Assistant Professor in the Department of Computer Science a San Diego State University. He received his Ph.D. in Computer Science from Louisiana State University-Baton Rouge in 2024. His research focuses on performance and scalability analysis of user facing web systems, stateful data streaming systems, and scientific data analytics systems in cloud and HPC environments. His work has appeared in systems conferences and journals including SoCC, Middleware (Best Paper Award), TPDS, IPDPS, and ICDCS. For more information, please visit https://jianshuliu1721.github.io/

HOST: Hajar Homayouni