[CS] Reminder - Varsha Rao MS Presentation/May 15, 2024

[Megan Woodward via cs]

This is an announcement of Varsha Rao's MS Presentation
===============================================
Candidate: Varsha Rao

Date: Wednesday, May 15, 2024

Time:  9:30 am CT

Location: JCL 298

Remote Location: https://uchicago.zoom.us/j/98092551443?pwd=M2VWT3FlbVBQQ3dGQjhGMW9QT2QvQT09<https://urldefense.com/v3/__https://uchicago.zoom.us/j/98092551443?pwd=M2VWT3FlbVBQQ3dGQjhGMW9QT2QvQT09__;!!BpyFHLRN4TMTrA!-0U_WnHbRXbvKacJx3mhKCazqHArIQv44JmndOCKQG751thxwO7mMtaIsYN0DhTOKb3OJfrmkKLERKi8vMgIYJw7WxICzVKXPxiQgUq6Pw$>

Meeting ID: 980 9255 1443
Passcode: 937540

Title: Characterizing the Opportunity for Reducing the Operational Carbon Footprint of Storage Systems

Abstract: Data center power consumption and its carbon emissions are increasing rapidly. Storage
systems are an important contributor to this problem. While storage technology and
management techniques have improved rapidly, the growth of storage capacity used has grown
faster, so improving efficiency is not enough. On the other hand, the efforts to decarbonize
power grids have increased the variability in average carbon intensity, creating an opportunity to
reduce carbon emissions by moving the flexible computing workload to low carbon periods.
We study the University of Chicago’s high energy physics storage system, to identify
opportunities to flex storage power consumption to reduce operational carbon footprint. First, we
build a workload model for the background tasks, characterizing their IO, compute and power
costs. Projecting to different data center scenarios shows these tasks consume ~2-4% of data
center power and ~12-21% of data center storage power annually. Projections for varying
hardware configuration show the shift to SSDs decreases the data center power consumption,
while the power consumption of background tasks increases relative to data center storage
power.
Second, we consider power grid ACI variation, studying several Independent System Operators
(ISOs) as settings. In one scenario, Hyperscale Data Center Worldwide (aggregate) in the
California Independent System Operator (CAISO), the background tasks' carbon emissions are
equivalent to the annual electricity use of ~323k US homes. By using carbon-optimized
scheduling policies, these emissions can be reduced by ~60%. Even greater reductions (up to
~81%) are possible as grids become more variable in the future. These results highlight the
potential of background tasks for reducing storage system emissions. Our study is the first to
show that data growth increases the power cost for data reliability, durability, availability and
efficient storage space utilization faster than overall storage system power cost.

Advisors: Andrew Chien

Committee Members: Andrew Chien, Haryadi Gunawi, and Sanjay Krishnan





-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cs.uchicago.edu/pipermail/cs/attachments/20240515/8e50e144/attachment-0001.html>