Enabling Fairness in Cloud Computing Infrastructures

Cloud computing has emerged as a key technology in many ways over the past few years, evidenced by the fact that 93% of the organizations is either running applications or experimenting with Infrastructure-as-a-Service (IaaS) cloud. Hence, to meet the demands of a large set of target audience, IaaS cloud service providers consolidate applications belonging to multiple tenants. However, consolidation of applications leads to performance interference with each other as these applications end up competing for the shared resources violating QoS of the executing tenants. This dissertation investigates the implications of interference in consolidated cloud computing environments to enable fairness in the execution of applications across tenants. In this context, this dissertation identifies three key issues in cloud computing infrastructures. We observe that tenants using IaaS public clouds share multi-core datacenter servers. In such a situation, we identify that the applications belonging to tenants might compete for shared architectural resources like Last Level Cache (LLC) and bandwidth to memory, slowing down the execution time of applications. This necessitates a need for a technique that can accurately estimate the slowdown in execution time caused due to multi-tenant execution. Such slowdown estimates can be used to bill tenants appropriately enabling fairness among tenants. For private datacenters, where performance degradation cannot be tolerated, it becomes critical to detect interference and investigate its root cause. Under such circumstances, there is a need for a real-time, lightweight and scalable mechanism that can detect performance degradation and identify the root cause resource which applications are contending for (I/O, network, CPU, Shared Cache). Finally, the advent of microservice computing environments, calls for a need to rethink resource management strategies in multi-tenant execution scenarios. Specifically, we observe that the visibility enabled by microservices execution framework can be exploited to achieve high throughput and resource utilization while still meeting Service Level Agreements (SLAs) in multi-tenant execution scenarios. To enable this, we propose techniques that can dynamically batch and reorder requests propagating through individual microservice stages within an application.PHDComputer Science & EngineeringUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttps://deepblue.lib.umich.edu/bitstream/2027.42/149844/1/ramsri_1.pd