Serverless · Kubernetes · Reinforcement Learning

Dr. Siddharth Agarwal

Software Engineer · Melbourne eResearch Group · University of Melbourne

I am a software engineer and cloud researcher with a PhD in serverless computing optimisation — focusing on cold start reduction, intelligent autoscaling, and dynamic function configuration. I currently build backend systems at the Melbourne eResearch Group, and my research applies reinforcement learning and ensemble methods to real-world cloud platforms including OpenFaaS and AWS Lambda.

Research

Contributions

Three interlocking problems in serverless resource management — each investigated empirically and validated against industry baselines.

Cold Start Frequency Reduction

RL

Static keep-alive policies cannot adapt to on-demand invocation patterns, causing unavoidable cold starts that spike tail latency.

  • Both on-policy (PPO) and off-policy (Q-learning) RL policies investigated
  • Validated on AWS Lambda & OpenFaaS under diverse workload traces

Significantly reduced cold start events vs. keep-warm & reactive baselines

Input-Aware Memory Configuration

ML

Fixed memory limits are chosen conservatively, systematically over-provisioning compute and inflating cost regardless of actual function workload.

  • Multi-output Random Forest trained on function input characteristics
57–87%
cost reduction vs. static allocation
98%
R² for execution time
96%
R² for memory prediction

Adaptive Autoscaling under Workload Uncertainty

RPPO

Threshold-based HPA cannot anticipate bursty or time-varying workloads, leading to under-scaling, SLA violations, and wasted idle replicas.

  • Recurrent PPO (RPPO) proposes on-policy scaling — outperforms DRQN baseline
22%
throughput improvement vs. Kubernetes HPA
37%
reduction in resource wastage
Software

Projects & Tools

Saarthi

AI-Powered

Intelligent serverless platform with workload prediction, mathematical optimisation, and automated resource management.

OpenFaaSPythonKubernetes

Serv-Drishti

Interactive Tool

Interactive visualisation engine for serverless computing internals — request routing, resource management, and performance analytics in real time.

JavaScriptChart.jsDocker

DRe-SCale

RL-Based

Proposes Recurrent PPO (RPPO) for intelligent serverless autoscaling. Outperforms DRQN and standard HPA — achieving 22% throughput improvement and 37% resource wastage reduction. Published at IEEE TSC 2024.

PythonKubernetesPyTorch

MemFigLess

Cost-Optimised

Input-aware dynamic memory configuration for serverless functions using Random Forest ensemble learning. Achieves 57–87% cost savings and 54–82% memory allocation reduction vs. static baselines on AWS Lambda.

AWS LambdaPythonDynamoDB

FaaSTrainGym

Serverless RL

Serverless reinforcement learning training platform integrating OpenFaaS with OpenAI Gym for scalable, distributed RL experiment pipelines.

OpenFaaSPythonPyTorch

Comprehensive 8-chapter deep dive into OpenFaaS internals: gateway architecture, watchdog, metrics, and practical implementations with code examples.

GoKubernetesDocker
Career

Experience

Oct 2026 — Present

Software Engineer

Melbourne eResearch Group · University of Melbourne

Building backend features and tests end to end on the Athena platform, a large-scale scheduling system.

Sep 2021 — Feb 2026

Graduate Research Fellow (PhD)

qCLOUDS Laboratory · University of Melbourne

Led 4+ end-to-end research projects. Designed Kubernetes clusters (5–20 nodes) on Melbourne Research Cloud (NeCTAR) and AWS. Built Saarthi, MemFigLess, DRe-SCale, and Serv-Drishti. Supervised by Prof. Rajkumar Buyya & Dr. Maria Rodriguez.

PythonAWSKubernetesOpenStack
Mar 2023 — Present

Sessional Academic — Head Tutor

University of Melbourne

Leading postgraduate tutorials for Distributed Systems (COMP90015, 100+ students/semester) and Distributed Algorithms (COMP90020). Also tutored Cluster & Cloud Computing, Advanced Database Systems, and Statistical Machine Learning.

Mar 2026 — Present

Sessional Academic

Monash University & RMIT University

Postgraduate cloud architecture, distributed systems, and cloud security tutorials at Monash. Weekly algorithm tutorials (25–30 students), assignment feedback, and curriculum collaboration at RMIT.

Aug 2020 — Jun 2021

Research Intern (Vocational Placement)

Telstra / University of Melbourne

Deployed serverless environments on Melbourne Research Cloud; investigated cold-start optimisation; delivered architectural recommendations to Telstra engineering stakeholders.

Apr 2018 — Jun 2019

Associate Systems Engineer

IBM India, Bengaluru

Developed and managed Archer CMS enterprise applications and SQL databases; led automation initiatives; client presentations and team knowledge transfer.

JavaSQL
Academic

Education

2021 — 2026

Doctor of Philosophy — Engineering and IT

University of Melbourne

Learning-Centric, Payload and QoS-Aware Function Resource Configuration and Management in Serverless Computing

  • Supervisors: Prof. Rajkumar Buyya & Dr. Maria Rodriguez
  • Melbourne Research Scholarships (fee offset + stipend) · Melbourne Welcome Grant (2022)
  • Engineering and IT Conference Travel Grant (2024)
  • Best Paper Award (Runner-Up), IEEE/ACM CCGrid 2021
2021 · Conferred 30 July 2021

Master of Science (Computer Science) with Distinction

University of Melbourne
  • WAM: 89.846 · Dean's Honours List, Faculty of Science (2021)
  • Supervisor: Prof. Rajkumar Buyya
2013 — 2017

B.Tech (Hons) in Computer Science & Engineering

Jaypee Institute of Information Technology, Noida, India
  • GPA: 8.5 / 10
Expertise

Skills & Technologies

Cloud & Orchestration

AWSGCPOpenStackKubernetesDockerOpenFaaS

Languages & DevOps

PythonJavaGoSQLAnsibleGitHub ActionsPrometheusGrafana

ML & Research

PyTorchScikit-learnNumPyDynamoDB
Get in Touch

Contact

I am open to research collaborations, academic discussions, and selected cloud consulting engagements. Feel free to reach out.

Stay Updated

Get notified when I publish new posts on cloud computing and distributed systems research.