[Colloquium] Reminder - Hai Nguyen Candidacy Exam/May 2, 2023

[Megan Woodward]

This is an announcement of Hai Nguyen's Candidacy Exam.
===============================================
Candidate: Hai Nguyen

Date: Tuesday, May 02, 2023

Time:  2:30 pm CST

Remote Location:  https://uchicago.zoom.us/j/94903341306?pwd=UWViUFM5WnNzREY0WmtBMTlFcW1mQT09<https://urldefense.com/v3/__https://uchicago.zoom.us/j/94903341306?pwd=UWViUFM5WnNzREY0WmtBMTlFcW1mQT09__;!!BpyFHLRN4TMTrA!8YNNtSW2zryOVCLpssArvincf1N6LSMJvViYXbE-WdNy90oisyeyz22MEMVQ_oGBC3M7lGMn9vwuMI2bWY8i9QyGxA$>

Location: JCL 298

Title: Rate-based Abstract Machine: An Efficient Computation Model for Guaranteeing Performance of Bursty, Real-time Applications

Abstract: Today's Function-as-a-Service (FaaS) systems only provide statistical guarantees, so bursty, real-time applications cannot rely on them to deliver quality guarantee computation. We propose Rate-based Abstract Machine (RBAM) -- an extension of the FaaS computation service that associates a guaranteed invocation rate to the FaaS function. This guarantee enables timely invocation allocation, enabling applications to meet real-time requirements.
We develop analytical models to study RBAM, proving its capability in bounding FaaS invocation allocation latency which is sufficient to guarantee applications' real-time deadlines. We also use the analytical model to prove the feasibility of RBAM implementation and derive a naive RBAM allocation algorithm. We use this algorithm to create an RBAM realization, named Real-time Serverless (RTS).
To demonstrate RBAM applicability, we implement a distributed real-time video analytic
application with Real-time Serverless. RTS ensures both statistical and absolute application
performance guarantees. These new capabilities are also robust against a wide range of workload burstiness, opening new performance engineering space with flexible deployments. Further, we use RTS as a building block to create Storm-RTS, a stream processing engine that utilizes the rate guarantee to implement application performance transparency and predictability. Such new features allow Storm-RTS to gain robust performance stability, high deployment flexibility, and can even optimize application deployment for different objectives with simple declarative policies.
Given this foundation, we will provide an efficient RBAM implementation. We plan to propose several guaranteed invocation rate allocation algorithms that can deploy any rate-guarantee with low overhead. We will also develop analytical studies and simulations to show their efficiency and then implement them inside RTS. We expect to make RTS implementation cost asymptotically small for a large number of RBAMs.

Advisors: Andrew Chien

Committee Members: Andrew Chien, Kyle Chard, Junchen Jiang, and Sanjay Krishnan




-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cs.uchicago.edu/pipermail/colloquium/attachments/20230502/1d4ea343/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Hai-candidacy-document.pdf
Type: application/pdf
Size: 3864874 bytes
Desc: Hai-candidacy-document.pdf
URL: <http://mailman.cs.uchicago.edu/pipermail/colloquium/attachments/20230502/1d4ea343/attachment-0001.pdf>