Characterizing Serverless Workloads: A Large-Scale Analysis of Function-as-a-Service Platforms
Saman Akbari, Yassine Lazreg, Mohammadreza Malekabbasi, Manfred Hauswirth
Useful, timely consolidation of eight public FaaS traces with open code, but descriptive, largely confirmatory, and with confounded trend claims and unvalidated recommendations.
Abstract
Serverless computing is a popular approach to application deployment without having to manage the underlying infrastructure. Its most common implementation is Function-as-a-Service (FaaS), in which applications are structured as fine-grained functions. FaaS providers face non-trivial tasks in the management of their platforms, including cold starts, auto-scaling, and scheduling. The design of optimized platforms requires an accurate understanding of realistic serverless workloads. However, existing research and platforms rely on insights from single, often outdated datasets. In this work, we conduct a large-scale, cross-platform comparative analysis of eight publicly available FaaS workload traces from major cloud providers (Alibaba, Azure, Huawei, IBM) and a scientific computing platform (Globus). We provide a comprehensive and up-to-date view of the serverless ecosystem by characterizing key dimensions such as invocation patterns, latency, cold start behavior, and resource utilization. Our analysis shows that although invocation patterns remain universal, execution times have shortened, shifting the performance bottleneck toward platform scheduling and dependency management. We also observe an architectural shift toward "Serverless 2.0" models, where concurrent, long-lived containers are increasingly utilized to bypass cold start penalties. Furthermore, we found resource over-provisioning and extreme skew in user request volumes. These findings provide actionable insights for designing more efficient FaaS platforms, informing improved approaches to targeted cold start mitigation, predictive resource management, and dependency-aware scheduling.
AI Impact Assessments
(1 model)Core Contribution
This is a cross-trace empirical characterization study. It analyzes eight public FaaS traces (Azure19/20/21, Alibaba21, Huawei23/25, Globus24, IBM26) along invocation patterns, burstiness, diurnal cycles, popularity skew, idle times, cold starts, latency and overhead, instances, resource usage and limits, runtimes, triggers, I/O, and per-user tenancy. Its main contribution is putting these previously siloed datasets side by side and distilling ten "insights" for platform design. The headline claims are:
The novelty is synthesis and breadth, not new methods or new data. The code is open-sourced, which adds practical value.
Methodological Rigor
The analysis is descriptive and uses standard statistics: CDFs, percentiles, Lorenz curves, CV, and Spearman correlation. It uses Dask for scale and applies reasonable cleaning (dropped rows reported in Table 2). Several weaknesses limit the strength of the inferences:
Potential Impact
The paper is useful as a reference and entry point for serverless systems researchers. It supplies a consolidated table of trace schemas and metrics, concrete numbers (cold start breakdowns, memory/CPU limit vs. usage, concurrency defaults), and an argument for multi-trace evaluation. That argument is the most actionable message: algorithms tuned on Azure19 may not transfer. The paper is likely to be cited in motivation sections and in work on keep-alive, pre-warming, autoscaling, and trace-based benchmarking. It will not change the field's direction. Most findings (diurnal patterns, skew, heavy tails, long-tail cold starts) confirm prior single-trace studies. The newer observations (IBM26 concurrency and minimum-scale behavior, timer-trigger share, millisecond execution) are mostly taken from the source trace papers. The "insights" are fairly generic recommendations (per-function keep-alive, right-sizing, dependency-aware scheduling) and are not validated by experiments or simulation.
Timeliness & Relevance
It is timely. Serverless research leans heavily on the aging Azure19 trace, and several new traces have appeared recently (Huawei25, IBM26, Globus24). A consolidated view is a real community need, and the "Serverless 2.0" framing is topical. The value depends on trace availability, and the paper is a snapshot that will age as new traces appear. It proposes a common trace format but does not deliver one.
Strengths
Limitations
Overall
This is a solid, useful, but incremental measurement and synthesis paper. It is likely to be a handy reference with moderate citation impact in the serverless and cloud systems community, but with limited conceptual novelty or surprise.
Generated Nov 4, 2026
Comparison History (0)
Useful, timely consolidation of eight public FaaS traces with open code, but descriptive, largely confirmatory, and with confounded trend claims and unvalidated recommendations.