The Networking Team of one of the largest telecommunications equipment makers in the world found that even profiling tools based on processor register events could shed no light on the causes of the poor performance of their high-throughput, time-critical base station application.
Poor cache-software interaction was identified as the main bottleneck, and a strategy for improving cache usage efficiency was then put into practice. The resulting improved cache performance meant more network users per base station, while maintaining the same bottom line value.
CriticalBlue's intervention focused on the short-term issue, but as the solution was deployed, it became clear that the telecom provider wanted to tap into similar speed-up opportunities over their whole codebase. At the customer's request, CriticalBlue started the development of an automatic data structure optimizer.
The telecom equipment maker initially selected CriticalBlue as a partner on a joint project because a more comprehensive approach was needed to achieve ad hoc cache-oriented profiling within the real-time constraints of their application. Data structures in the application that were misusing the L1 cache were finally exposed, making room for a major performance breakthrough.