Beforehand, Meta used a mix of warmup (to get gamers prepared) and prefetch (to cache content material on disk) for video supply. Whereas these strategies helped enhance community effectivity, they launched vital challenges. Warmup required instantiating a number of participant situations sequentially, which consumed vital reminiscence and restricted preloading to only some movies. This excessive useful resource demand meant {that a} extra scalable sturdy resolution may very well be utilized to ship the moment playback anticipated on trendy, fast-scrolling social feeds.
Integrating Media3 PreloadManager
To attain really immediate playback, Meta’s Media Basis Shopper workforce built-in the Jetpack Media3 PreloadManager into Fb and Instagram. They selected the DefaultPreloadManager to unify their preloading and playback methods. This integration required refactoring Meta’s present structure to allow environment friendly useful resource sharing between the PreloadManager and ExoPlayer situations. This strategic shift offered a key architectural benefit: the power to parallelize preloading duties and handle many movies utilizing a single participant occasion. This dramatically elevated preloading capability whereas eliminating the excessive reminiscence complexities of their earlier method.
Optimization and Efficiency Tuning
The workforce then carried out intensive testing and iterations to optimize efficiency throughout Meta’s numerous international machine ecosystem. Preliminary aggressive preloading generally prompted points, together with elevated reminiscence utilization and scroll efficiency slowdowns. To resolve this, they fine-tuned the implementation through the use of cautious reminiscence measurements, contemplating machine fragmentation, and tailoring the system to particular UI patterns.
Wonderful tuning implementation to particular UI patterns
Meta utilized totally different preloading methods and tailor-made the conduct to match the precise UI patterns of every app:
Fb Newsfeed: The UI prioritizes the video presently coming into view. The supervisor preloads solely the present video to make sure it begins the second the person pauses their scroll. This “current-only” focus minimizes information and reminiscence footprints in an surroundings the place customers may even see many static posts between movies. Whereas the system is presently designed to preload simply the video in view, it may be adjusted to additionally preload upcoming (future) movies.
Instagram Reels: This can be a pure video surroundings the place customers swipe vertically. For this UI, the workforce applied an “adjoining preload” technique. The PreloadManager retains the movies instantly after the present Reel prepared in reminiscence. This bi-directional method ensures that whether or not a person swipes up or down, the transition stays immediate and clean. The end result was a dramatic enchancment within the High quality of Expertise (QoE) together with enhancements in Playback Begin and Time to First Body for the person.
Scaling for a various international machine ecosystem
Scaling a high-performance video stack throughout billions of units requires extra than simply aggressive preloading; it requires intelligence. Meta confronted preliminary challenges with reminiscence stress and scroll lag, notably on mid-to-low-end {hardware}. To resolve this, they constructed a Gadget Stress Detection system across the Media3 implementation. The apps now monitor I/O and CPU alerts in real-time. If a tool is beneath heavy load, preloading is paused to prioritize UI responsiveness.
This device-aware optimization ensures that the advantage of immediate playback would not come at the price of system stability, permitting even customers on older {hardware} to expertise a smoother, uninterrupted feed.
Architectural wins and code well being
Past the user-facing metrics, the migration to Media3 PreloadManagerprovided long-term architectural advantages. Whereas the combination and tuning course of wanted a number of iterations to steadiness efficiency, the ensuing codebase is extra maintainable. The workforce discovered that the PreloadManager API built-in cleanly with the prevailing Media3 ecosystem, permitting for higher useful resource sharing. For Meta, the adoption of Media3 PreloadManager was a strategic funding in the way forward for video consumption.
By adopting preloading and including device-intelligent gates, they efficiently elevated whole watch time on their apps and improved the general engagement of their international group.
Ensuing impression on Instagram and Fb
The proactive structure delivered fast and measurable enhancements throughout each platforms.
Fb skilled sooner playback begins, decreased playback stall charges and a discount in unhealthy periods (like rebuffering, delayed begin time, decrease high quality,and many others) which total resulted in increased watch time.
Instagram noticed sooner playback begins and a rise in whole watch time. Eliminating be a part of latency (the interval from the person’s motion to the primary body show) straight elevated engagement metrics. The fewer interruptions as a consequence of decreased buffering meant customers watched extra content material, which confirmed via engagement metrics.
Key engineering learnings at scale
As media consumption habits evolve, the demand for fast experiences will proceed to develop. Implementing proactive reminiscence administration and optimizing for scale and machine variety ensures your utility can meet these expectations effectively.
Give attention to delivering a dependable expertise by minimizing stutters and loading instances via preloading. Reasonably than easy disk caching, leveraging memory-level preloading ensures that content material is prepared the second a person interacts with it.
Customise preloading conduct as per your apps’s UI. For instance, use a “current-only” focus for combined feeds like Fb to avoid wasting reminiscence, and an “adjoining preload” technique for vertical environments like Instagram Reels.
Integrating with Media3 APIs reasonably than a customized caching resolution permits for higher useful resource sharing between the participant and the PreloadManager, enabling you to handle a number of movies with a single participant occasion. This leads to a future-proof codebase that’s simpler for engineering groups to not solely preserve and optimize over time but additionally profit from the most recent characteristic updates.
Broaden your market attain by testing on varied units, together with mid-to-low-end fashions. Use real-time alerts like CPU, reminiscence, and I/O to adapt options and useful resource utilization dynamically.
Integrating with Media3 APIs reasonably than a customized caching resolution permits for higher useful resource sharing between the participant and the PreloadManager, enabling you to handle a number of movies with a single participant occasion. This leads to a future-proof codebase that’s simpler for engineering groups to not solely preserve and optimize over time but additionally profit from the most recent characteristic updates.
Broaden your market attain by testing on varied units, together with mid-to-low-end fashions. Use real-time alerts like CPU, reminiscence, and I/O to adapt options and useful resource utilization dynamically.