Agenda
Solving Memory Issues at Scale: Anvil's Memory Analytics Pipeline
Vladimir Katchadourian (Senior Data Scientist, Ubisoft)
Pass Type:
Academic Pass, Festival Pass, Game Changer Pass, Indie Pass, Start-Up Pass -
Get your pass now!
Track:
Programming
Audience Level:
All
Format:
Lecture
Vault Recording: TBD
Audience Level: All
This talk presents the methodology and practical solutions behind Ubisoft's end-to-end memory investigation workflow developed for Anvil. By recording and processing every allocation and deallocation across multi-hours game sessions into compact files and pairing it with a data analytics pipeline, the approach enables build-over-build regression tracking and interactive investigation. The presenter shows how this workflow addresses the full range of real-world memory problems: detecting leaks and regressions, correlating allocation spikes with frame-time regressions, monitoring memory budgets and asset costs or improving custom allocation strategies.
He explains how combining local-first processing and interactive analysis with large-scale analytics empowers everyone, from engine programmers to technical artists, at a negligible cost.
He explains how combining local-first processing and interactive analysis with large-scale analytics empowers everyone, from engine programmers to technical artists, at a negligible cost.
Takeaway
Attendees will learn how a unified memory profiling pipeline addresses the full spectrum of AAA memory related problems, from low-level issues to production health monitoring and investigation, and how it dramatically prevents and simplifies investigations and reduces production costs by relying on a local-compute-first data pipeline architecture.
Intended Audience
This lecture targets engine programmers, gameplay programmers, technical directors, technical artists, and technical producers working on AAA or large-scale games. Basic familiarity with memory management concepts is helpful but not required; the tooling and methodology are presented from the ground up.