Liam Arzola

I’m a PhD student at UCSD’s Systems and Networking group, where I’m advised by Amy Ousterhout and Alex Snoeren. My current research focuses on μs scale scheduling for hardware accelerators.

Before joining UCSD, I was a software engineer at Microsoft in the Core OS>Virtual Machines and Containers>Linux on Windows team, located in Seattle. Before that I was a research intern at the MPI-SWS Operating Systems Group, where I was advised by Antoine Kaufmann and worked on investigating TCP stacks for virtualized environments and then for μs scale reconfigurable optical networks. I completed my Bachelor’s degree at Cornell University where I conducted research (ongoing) under the supervision of Lorenzo Alvisi investigating byzantine fault tolerant relational databases.

I’m excited about computer systems, construed broadly, with an interest in tackling problems concerning the fault tolerance and performance of operating systems, networks, distributed systems, and computer architectures. Based on my past research experiences, I find two domains particularly fascinating: datacenter systems and networks, and byzantine fault tolerant systems.

selected publications

2023

Virtuoso: High Resource Utilization and μs-scale Performance Isolation in a Shared Virtual Machine TCP Network Stack

Matheus Stolet, Liam Arzola, Simon Peter, and Antoine Kaufmann

Under Submission, 2023

Abs PDF

Virtualization improves resource efficiency and ensures security and performance isolation for cloud applications. To that end, operators today use a layered architecture that runs a separate network stack instance in each VM and container connected to a separate virtual switch. Decoupling through layering reduces complexity, but induces performance and resource overheads that are at odds with increasing demands for network bandwidth, communication requirements for large distributed applications, and low latency. We present Virtuoso, a new software networking stack for VMs and containers. Virtuoso performs a fundamental re-organization of the networking stack to maximize CPU utilization, enforce isolation, and minimize networking stack overheads. We maximize utilization by running one elastically shared network stack instance on dedicated cores; we enforce isolation by performing central and fine-grained per-packet resource accounting and scheduling; we reduce overheads by building a single-layer data path with a one-shot fast-path incorporating all processing from the TCP transport layer through network virtualization and virtual switching. Virtuoso improves resource utilization by up to 50%, latencies by up to 42% compared to other virtualized network stacks without sacrificing isolation, and keeps processing overhead within 11.5% of unvirtualized network stacks.