Attributing the Unseen: High-Precision CPU Profiling via Trace-Join Analytics
Fuente:
Zenodo
Enregistré dans:
| Auteur principal: | |
|---|---|
| Format: | Recurso digital |
| Publié: |
Zenodo
2026
|
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866901496079056896 |
|---|---|
| author | Varun Raj |
| author_facet | Varun Raj |
| contents | <p>One of the major challenges of modern distributed systems is to understand CPU consumption patterns. Existing profiling tools cannot indicate which experiments or traffic patterns caused the CPU to be busy. This creates challenges for developers in diagnosing performance issues. The proposed system uses unique sequence tagging to correlate profiling with attribution. Each profile is tagged with the id of the associated request, which is persisted through the profile's execution, and is joined at the batch analytics stage with the request metadata, including the experiment identifiers, the latency classifications and the traffic slice identification. Developers can query the dataset for profiles matching their attribution settings, then visualize the result in a flame graph. The flame graph is a method of visualizing profiles in an easy to understand manner, displaying what paths through code are consuming too much CPU in the context of an experiment. By enabling attribution-based profiling, moved performance debugging efforts from relying on infrastructure engineers and experts to be self-serve by product developers. As a result, performance regressions can be detected earlier in the development cycle and their occurrence prevented, saving computational resources and protecting system stability in production environments serving global traffic. The same attribution can be applied on other dimensions, such as memory or network profiling</p> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_19163536 |
| institution | Zenodo |
| language | |
| publishDate | 2026 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | Attributing the Unseen: High-Precision CPU Profiling via Trace-Join Analytics Varun Raj <p>One of the major challenges of modern distributed systems is to understand CPU consumption patterns. Existing profiling tools cannot indicate which experiments or traffic patterns caused the CPU to be busy. This creates challenges for developers in diagnosing performance issues. The proposed system uses unique sequence tagging to correlate profiling with attribution. Each profile is tagged with the id of the associated request, which is persisted through the profile's execution, and is joined at the batch analytics stage with the request metadata, including the experiment identifiers, the latency classifications and the traffic slice identification. Developers can query the dataset for profiles matching their attribution settings, then visualize the result in a flame graph. The flame graph is a method of visualizing profiles in an easy to understand manner, displaying what paths through code are consuming too much CPU in the context of an experiment. By enabling attribution-based profiling, moved performance debugging efforts from relying on infrastructure engineers and experts to be self-serve by product developers. As a result, performance regressions can be detected earlier in the development cycle and their occurrence prevented, saving computational resources and protecting system stability in production environments serving global traffic. The same attribution can be applied on other dimensions, such as memory or network profiling</p> |
| title | Attributing the Unseen: High-Precision CPU Profiling via Trace-Join Analytics |
| url | https://doi.org/10.5281/zenodo.19163536 |