CAREER: Hardware/Software Co-design for Serverless Computing

This award is funded in part under the American Rescue Plan Act of 2021 (Public Law 117-2).

Cloud computing has become a dominant computing paradigm in diverse sectors. Behind this considerable success, decades of research efforts have evolved cloud environments to offer better services to their users. Keeping up with this trend, the current cloud community is adopting a new cloud-computing paradigm, serverless computing, which allows cloud users to have conceptual distance from the underlying hardware resources and focus instead on developing their applications.

Despite its benefits, the serverless platform raises new challenges. One of the apparent problems is slow service time. The serverless platform needs to provide as timely responses to users' requests as other cloud applications.

However, its high dependency on the network causes long service latency, significantly hurting users' experiences. This project addresses this slow service time by reducing dependence on network memory resources using emerging devices, namely, smart network interface cards (SmartNIC) and non-volatile memories (NVM). The successful outcomes directly benefit cloud-computing technology, one of the crucial areas in modern computer technology.

This improvement leads to a better cloud-computing environment, helping cloud users be more productive with serverless computing but without concerns about its slow service time. This project also focuses on educating computer science students on hardware and nurturing them to be advanced engineers capable of designing hardware/software co-designed systems.

The education plans include diverse activities for female and under-represented-minority students, providing them with backgrounds in computer architecture.

Rather than investigating only software optimizations, which often offer expensive trade-offs, this project plans to redesign both the software and hardware platform to address serverless platforms' slow service latency. Specific research objectives to achieve this goal include: (1) migrate the entire storage protocols from the host kernel to the SmartNIC; (2) implement a non-volatile cache controller, which inserts a non-volatile cache layer between the main memory and the network storage; (3) employ non-volatile memory to enhance the non-volatile cache.

This project dissects the network storage latency into two sub-problems, network latency and storage access latency. The first research objective addresses network latency by offloading protocol processing burdens from the host CPU. The other two objectives address slow storage latencies by adopting the SmartNIC-controlled non-volatile cache with the caching space in the local disks and the NVMM.

The proposed design will be one of the references for the next-generation cloud server, showing how the hardware and the software can be co-designed to satisfy the high software demands.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.