PaPeR
PaPeR: AWS Distributed Video Benchmarking
Distributed Video Compression Infrastructure
High-throughput distributed cloud infrastructure built on AWS with Terraform, collaborating with Meta, Intel, and SPIE to automate video compression quality evaluation across thousands of parallel transcoding jobs.
The Core Challenge
Engineering Solution & Implementation
Engineered an automated cloud pipeline using Flask, AWS EC2, S3, RDS, Lambda, and modular Terraform infrastructure. Provided instant containerized test runners that automatically execute transcoding experiments, collect PSNR/VMAF metrics, and output standardized SPIE evaluation datasets.
Figure 5: PaPeR's distributed AWS cloud architecture for automated parallel video quality benchmarking.
The Flask control plane routes an experiment to Terraform-provisioned runners and tracks session state in MySQL. Runners validate quality with PSNR and VMAF and export SPIE datasets. SPIE experimental setup time stays separate from infrastructure provisioning, and the session record follows a batch from input through metric export.
My contribution
I engineered the AWS and Terraform pipeline: Flask session control, containerized runners, and automated PSNR and VMAF collection for the distributed video benchmarks delivered to Meta.
Tradeoff
The 99% reduction is SPIE experimental setup time, from 3 hours to seconds. It is separate from the 75% reduction in infrastructure provisioning, the 50% increase in session throughput, and the 1,800+ hours across 600+ experiments.
Verified Outcomes & Deliverables
Achieved a 99% reduction in SPIE experimental setup time on AWS, from 3 hours to seconds.
Reduced infrastructure provisioning time by 75%.
Increased session processing throughput by 50%.
Saved over 1,800 engineering hours across 600+ large-scale distributed video experiments.
Accelerated integration of V-Nova video compression technologies into Meta services (WhatsApp, Instagram, XR platforms).