SREcon26 Americas - Taming the Unpredictable: Reliability in Chaos @UsenixOrg

Uploaded April 2026 | Updated September 2026, 3 weeks ago

Taming the Unpredictable: Reliability in Chaos

Michelle Brush, Google Inc.

The increasing volume of code written by AI will only accelerate the complexity of our systems. We are moving beyond predictable systems that can be managed with traditional methods like thorough project plans, runbooks, and unit tests, into an era of truly complex systems that are vast and difficult to comprehend fully. These immensely complex systems will behave almost nondeterministically. We need new strategies.

This presentation will delve into why robust reliability practices are not just helpful but essential for navigating this explosion in complexity. It will share strategies for conquering unpredictability including building rigorous evaluations, implementing generic mitigations, recreating reproducibility, and developing software with a risk-first approach. Most importantly, it will discuss why humans will always be critical to taming the ever-growing complexity.

View the full SREcon26 Americas program at usenix.org/conference/srecon26americas/program

SREcon26 Americas - Taming the Unpredictable: Reliability in Chaos

NSDI 26 - RollPacker: Taming Long-Tail Rollouts for RL Post-Training with Tail Batching

NSDI 26 - KeepON: Supporting Deterministic Traffic on Standard NICs

NSDI 26 - FENIX: Enabling In-Network DNN Inference with FPGA-Enhanced Programmable Switches

Computer Security and Voting, Invited Talk by David Dill at USENIX Security 07

PEPR 26 - Toward Provably Private Insights into AI Use

PEPR 26 - DPSynth: From Research to Production—Engineering Differentially Private Synthetic...

$NSDI 26 - FRCC: Towards Provably Fair and Robust Congestion Control FRCC: Towards Provably Fair and Robust Congestion Control Anup Agarwal, Carnegie Mellon University; Venkat Arun, University of Texas at Austin; Srinivasan Seshan, Carnegie Mellon University Congestion control algorithms (CCAs) play a critical role in network bandwidth allocation. Recent work (from SIGCOMM 2022) showed that a large class of CCAs, including BBR, Copa, and Reno, starve flows in the presence of network jitter. Starvation occurs because CCAs coordinate fairness by encoding fair rates into congestion signals. For example, Renos throughput scales as 1/√loss rate. Even a small amount of noise in these signals leads to large errors in inferring fair rates. We present FRCC (Fair and Robust Congestion Controller), the first CCA that provably bounds unfairness (avoids starvation) even under network jitter. Our key insight is to encode only the flow count (or equivalently, the fair link fraction) into the congestion signals, and independently estimate the link capacity to calculate the fair rate. In this way, we bound jitters impact on fairness. We implement FRCC in the Linux kernel and evaluate it in a variety of network conditions, including synthetic jitter, heterogeneous RTTs, and multi-bottleneck settings. FRCC closely matches the bounds predicted by our theoretical analysis, and consistently achieves fairness, even when state-of-the-art CCAs exhibit starvation. View the full NSDI 26 program at https://www.usenix.org/conference/nsdi26/technical-sessions$

USENIX Security 25 - ALERT: Machine Learning-Enhanced Risk Estimation for Databases Supporting...

NSDI 26 - cc-pipe: Breaking Systemic Bottlenecks in RPKI Data Supply Chain with Concurrent and...

NSDI 26 - Who Watches the Watchers? On the Reliability of Softwarizing Cloud Application Management

NSDI 26 - SLATE: Service Layer Traffic Engineering