ScaleLens detects 4.2x more scalability faults in large distributed systems

A team led by Hao-Nan Zhu at UC Davis published the first comprehensive study on scalability faults in large-scale distributed systems. They manually reviewed 444 issue reports from 10 production systems (including Cassandra, HDFS, and Ignite) and found that most scalability faults arise from a synergy between dimensional code fragments (code whose cost grows with scale) and anti-patterns (e.g., O(n²) operations hidden in loops). The study classifies these patterns into root cause categories, giving developers a taxonomy to anticipate failures before deployment.

Based on these findings, the team designed ScaleLens, a detection tool that fuses dynamic analysis (to identify hot code paths) with static analysis (to match code against anti-pattern templates). In evaluations, ScaleLens outperformed existing tools by 4.2x in finding dimensional code fragments associated with known faults. When run on the latest stable versions of Cassandra, HDFS, and Ignite, it flagged 334 fragments with confirmed problematic behavior. The tool can be integrated into CI pipelines, offering a proactive way to catch latent scalability bugs before they crash production environments at petabyte scale.