Pranav Bhandari PhD Student at Emory University Storage, Caching, Workload Characterization

About

I am a PhD student in the Computer Science and Informatics program at Emory University. I work on problems related to workload characterization in storage and caching systems with Dr. Avani Wildani as a part of the Emory Simbiosys Lab. This is my CV.

Research

A system based on the workload makes efficient use of its resources based on the optimization goal (cache hit rate, energy, message traffic). But how do we evaluate workloads? Are qualitative descriptions like ‘write heavy’, ‘large sequential reads’ enough? No, they are not. Workloads are dynamic and complicated; therefore, we need a set of quantifiable features that can help in system design. Here are some of the questions that I think about most in my research.

  • Workload Characterization: What features of the workload influnce the optimization goal? How important is each feature? How to identify workload phase shifts?
  • System Design and Tuning: How to design a system for a given workload? How to react to change in workload? What is the cost of tuning? How often should a system be evaluated and tuned?

Filesystem Cache Tuning

Filesystems have default cache parameter but the general parameters are not optimal for all workloads. We analyze application I/O pattern in order to tune the cache configurations of a filesystem. We use traces from applications running on IBM GPFS to tune pagepool parameters. Pagepool is the GPFS page cache. It does not use the Linux page cache. Our goal is to develop a tool that can make filesystem cache recommendation based on periodic sampling of the workload.

Multi-Tier Caching

With the availability of multiple devices for caching, multi-tier solutions have more potential than ever. For a given workload, there can be various configurations that would be ideal depending on the performance and cost constraints. Our goal is to develop a framework for workload characterization of multi-tier caching workloads in order to design workload aware multi-tier cache systems.

Projects

I/O Workload Classification using CNN

We convert block I/O into access pattern images. We classify these images based on the labels which is determined by the source of the data. We show that workload labels can mean different things depending on the dataset. We analyze images from the CNN layers in order to analyze the features being used for classification.

PDF

PyMimircache

PyMimircache is an open source cache simulation framework developed by Junchen Yang as part of the Emory SimBioSys Lab. I implemented miniature-simulations of a workload based on paper in FAST'15: Efficient MRC Construction with SHARDS by Carl A. Waldspurger, Nohhyun Park, Alexander Garthwaite, and Irfan Ahmad, CloudPhysics, Inc.

GitHub

akamaiSimulator

A prototype multi-tier cache simulator using PyMimircache developed as a part of the smart cache project by Junchen Yang. I implemented the check for testing if a hit in L1 would have been a hit in L2 as well to determine the L1/L2 split.

GitHub

Multi-Tier Cache Simulator

I am working on a multi-tier simulation framework based on PyMimircache. The goal is to be able to simulate and experiment with different additional complexities having multiple tiers in the cache brings such as data placement, replacement policies, devices and their characteristics at each layer.

Using Smart Agents to Improve Connectivity in a Segmented Multi-Radio Wireless Network using SDNs

The issue of scalability is one of the biggest problems in the area of wireless mesh networking hindering its wide adoption despite the obvious benefits. Being an inexpensive approach compared to other network model, wireless mesh network is a perfect tool to connect underprivileged areas to the global network. We introduce a new component in our network model called the smart agent which has a bird’s eye view of the network and we use this information to implement policies that influence the connection of the second interface of multi-radio nodes.

PDF

Publications

Desperately Seeking ... Optimal Multi-Tier Cache Configurations

Tyler Estro, Pranav Bhandari, Avani Wildani, Erez Zadok

12th {USENIX} Workshop on Hot Topics in Storage and File Systems (HotStorage 20)

PDF

Shuffled Frog Leaping Algorithm for 0/1 Knapsack Problem on the GPU

Pranav Bhandari, Rahul Chandrashekhar, Peter Yoon

CSC'15 - The 2015 International Conference on Scientific Computing, Las Vegas, Nevada

PDF

Bio

Aug 2017 - Research Assistant
Department of Computer Science, Emory University
Jan 2018 - May 2019 Teaching Assistant
CS170 - Introduction to Computer Science (Spring 2018)
CS323 - Algorithms (Fall 2018, Spring 2019)
Department of Computer Science, Emory University
Jan 2016 - Jan 2017 Software Engineering Intern
CivicLift
Oct 2015 - Feb 2016 Student Apprentice
Independent Software
May 2013 - May 2014 Research Assistant
Department of Computer Science, Trinity College

Education

Doctor of Philosophy in Computer Science
2013 - 2017 Trinity College
Bachelor of Science in Computer Science and Mathematics

Contact

400 Dowman Drive
Department of Computer Science
Emory University
Atlanta, GA 30322