Akshitha Sriraman, PhD candidate in CSE, has received a 2020 Facebook Fellowship to support her work enabling more efficient hyperscale web services. The award recognizes promising doctoral students engaged in innovative and relevant research, with just over a 1% acceptance rate for applicants. In addition to tuition and financial support, fellowship recipients receive conference travel support and a paid visit to Facebook headquarters for the annual Fellowship Summit.
Sriraman works with advisor Prof. Thomas F. Wenisch to enable hyperscale computing on web services, which gives high-demand systems the ability to scale with large increases in traffic. Services like social media, web search, and content recommendations have high performance expectations from users, while the service providers try to meet this demand as cheaply and efficiently as possible.
“Unfortunately,” says Sriraman, “modern hyperscale web systems introduce trade-offs between performance and numerous features essential for cost- and energy-efficient operation of data centers.” Her research works to reconcile the performance vs. cost and energy efficiency conflict in datacenters through scalable solutions across the system stack.
Sriraman’s solutions to these problems lie at the intersection of architecture and software systems. Her solutions have improved the performance, cost, and energy efficiency of modern hyperscale datacenter systems and been put to use by major companies:
- The idea of searching the hardware design space to tune existing hardware unit types to better support an assigned web service. This helps avoid the expense of installing a number of specialized hardware components. A team of Facebook engineers has exploited her strategy to boost performance and cut down on cost at scale. Additionally, Facebook partnered with hardware vendords to design the next server generation based on her conclusions.
- An open-source software system that automatically configures threading to improve service performance. Microsoft is implementing insights from this system to improve service performance.
- Analytical models for hardware acceleration that project service speedup. These models then allowed her to project gains from accelerating key repeated operations identified by her characterization. This helps avoid investment in accelerators that ultimately won’t have a large impact on overall datacenter performance.
Previously, Sriraman’s research has appeared at top-tier systems venues like OSDI, ISCA, and ASPLOS. She has also been recognized with a Rackham Merit Ph.D. Fellowship and was selected for the Rising Stars in the EECS 2019 workshop.