Incident management · Production

Netflix formulates out-of-memory kill prediction on streaming devices as a machine learning classification problem

The problem

TVs and set-top boxes running Netflix have tighter memory constraints than compute devices, making out-of-memory kills a common cause of app crashes. Netflix needed a way to predict OOM kills in advance so it could take pre-emptive device-specific actions to avoid crashing.

Workflow diagram · grounded in source

Device and runtime data collection

integration

“The device capability data needed for the prediction was distributed in over three different schemas across the Big Data Platform. Joining them together and building a single indexable schema that can directly become a part of a bigger d…”

Graded window labeling

validation

“we label the memory readings of the sessions in a fixed window before the OOM kill as erroneous, and the rest of the entries as non-erroneous. In order to make the labeling more granular, and bring more variation in a binary classificati…”

Feature construction

ai_action

“The dataset now consists of several entries — each of which has certain runtime features (navigational level and memory reading in our case) and device characteristics (a mix of over 15 features that may be numerical, boolean or categori…”

Multi-class OOM classification

ai_action

“Now we can use any multi-class classification algorithm — ANNs, XGBoost, AdaBoost, ElasticNet with softmax etc.”

Pre-emptive device action

output

“we can take device specific actions to pre-emptively lower the performance in favor of not crashing — aiming to give the user the ultimate Netflix Experience”

Reported outcome

The article describes the methodology for formulating OOM kill prediction as an ML classification problem, covering dataset curation, labeling strategy, and feature engineering.
Actual model results and confusion matrices were redacted for confidentiality.

Reported metrics

Non-kill entries in datasetover 99.1%

OOM kill entry rate in dataset0.9%

Reported stack

Big Data PlatformANNsXGBoostAdaBoostElasticNet

Source

https://netflixtechblog.com/formulating-out-of-memory-kill-prediction-on-the-netflix-app-as-a-machine-learning-problem-989599029109

Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

The article describes the methodology for formulating OOM kill prediction as an ML classification problem, covering dataset curation, labeling strategy, and feature engineering.

What tools did this team use?

Big Data Platform, ANNs, XGBoost, AdaBoost, ElasticNet.

What results were reported?

Non-kill entries in dataset: over 99.1%; OOM kill entry rate in dataset: 0.9% (source-reported, not independently verified).

How is this incident management AI workflow structured?

Device and runtime data collection → Graded window labeling → Feature construction → Multi-class OOM classification → Pre-emptive device action.

The article describes the methodology for formulating OOM kill prediction as an ML classification problem, covering dataset curation, labeling strategy, and feature engineering.Actual model results and confusion matrices were redacted for confidentiality.