Scaling Performance Issue Detection and Diagnosis in Cloud Infrastructures

Cloud infrastructures today consist of heterogeneous software and hardware environments to serve different workloads. Efficiently and accurately analyzing the performance for these systems is of critical need. However, current performance analysis methods scale poorly to handle the performance data from modern cloud platforms and do not consider the effect of system heterogeneity. In this paper, we present SDiag, a scalable performance analysis tool to uncover performance issues in complex heterogeneous cloud platform. Specifically, SDiag first breaks down tracing records into different buckets. SDiag then analyzes and compares different buckets to find the performance bottleneck and critical path. Finally, SDiag provides supporting evidence to help developers effectively troubleshoot the reported performance issues.