Function-as-a-Service (FaaS) platforms enable easy deployment and hosting of code known as serverless functions and have gained considerable traction among software developers. Recently, Amazon has offered ARM64 processors as an alternative to x86 for hosting serverless functions on AWS Lambda. To encourage developers to migrate code to ARM64, use of ARM64 processors has been incentivized with a 20% discount. In this paper, we investigate the implications of adopting ARM64 processors for hosting serverless functions. We deploy and benchmark 18 distinct serverless functions that stress a variety of system resources. Additionally, we scaled up the runtime for each of our functions by increasing the work they performed using 40 incremental steps to increase runtime from a few seconds to several minutes. We compared performance differences of x86 vs. ARM64 processors for our serverless function workoads and present our findings summarizing differences in: CPU utilization, function runtime, function runtime variation, and hosting costs. While only 7 of 18 functions ran faster on ARM64, 15 of 18 were less expensive thanks to the cloud provider discount. Performance variation on ARM64 processors was less than half compared to x86 processors potentially from the lack of hyperthreading and lower resource contention for ARM64 CPUs.
Serverless computing simplifies application deployment by removing the need for infrastructure management, with RESTful APIs being the common interface. However, REST can lead to inefficiencies such as data over-fetching and under-fetching, which impact performance and cost. This paper investigates GraphQL as an alternative to REST for serverless functions using a serverless image processing pipeline. We evaluate roundtrip time (RTT), scalability, and cost, while also examining managed (AWS AppSync) and unmanaged (Apollo Server) GraphQL hosting solutions. Our results show that GraphQL generally outperforms REST with respect to pipeline RTT, especially when there is high network latency, offering a potentially better fit for optimizing data transfer in serverless applications.
Sky Computing creates an all-in-one computing platform using resources from various cloud providers. A Serverless Sky architecture can aggregate serverless resources to achieve low costs, high performance, and fault tolerance. This research will create a Serverless Sky platform to then: evaluate performance implications, design and evaluate different architectures, and investigate autonomous resource aggregation for smart self-management.