24 hours of profiling on one page

Recently, Odnoklassniki started profiling Java services in the cloud 24/7. This means about 40,000 applications profiled simultaneously. A task of this scale requires a set of solutions: write your own profiler, known as async-profiler, wrap it with an eBPF program to improve performance within your own cloud, continuously save the results to your own S3-compliant storage implementation. 

All of this has been covered many times before, but in this talk, the speaker will elaborate on his little brick - the problem of processing and displaying this data. How exactly from an algorithmic point of view to compress gigabytes of profile into a single HTML page without losing meaningful data, as well as how to turn JS inside out from an engineering point of view so that the browser can acceptably handle such volumes.