logo EECS Rising Stars 2023




Sara McAllister

Efficient and Sustainable Data Retrieval Systems



Research Abstract:

Today, the web contains zettabytes of data, mostly stored in massive, hyperscale datacenters. These datacenters enable much of today's modern computing from machine learning to social media. However, datacenters are projected to account for 33% of the global carbon emissions by 2050. While most datacenters are moving to renewable energy, the majority of their emissions are not from energy generation, but rather from embodied emissions, generated from lifecycle events like manufacturing and mining raw materials. One key way to reduce embodied emissions is to increase device lifetime. Unfortunately, extending the lifetime of storage is challenging because storage hardware becomes less reliable with both lifetime and IO accesses. My research focuses on how to reduce IO to enable longer storage-device lifetimes, thus enabling sustainable datacenter storage. In particularly, flash SSDs are limited to a small number of writes for their entire lifetime. For write-intense applications like flash caches, this write limit severely prohibits extending the flash SSD's lifetime. In my work, I designed flash caches that first tackle cache-level extraneous writes for small objects without large DRAM overheads and then address device-level write overheads by leveraging a new classification of flash interfaces that enable eliminating previously uncontrollable writes. Thus, my research enables sustainable flash caches through extending flash lifetimes even for the most difficult workloads.

Bio:

Sara McAllister is a PhD student at Carnegie Mellon University, advised by Nathan Beckmann and Greg Ganger. She is interested in computer systems, particularly caching and storage systems. Her work includes a focus on improving efficiency and sustainability through hardware-software co-design and grounding design choices in mathematical modeling. Her work has appeared at OSDI and SOSP, including receiving a Best Paper Award at SOSP 2021 for her paper "Kangaroo: Caching Billions of Tiny Objects on Flash". Sara also strives to increase inclusion in computer science, including by creating a DEI course for CS PhD students. Due to these efforts, she was awarded CMU's Graduate Student Service Award in 2022 and a Best Paper Award at SIGCSE 2023.