Bug 1626837 introduces a new memory allocation scheme in the profiler (which will only start being used later on, see meta bug 1612356). Thanks to off-thread allocation handling, the overhead seemed low in some early manual testing, but it would be good to get more confidence (and also include potential overhead in profiles, so the user is aware of profiler-related overhead), and later fix issues if any. To be recorded: - Time between request of a new chunk, and when the request gets handled. - Time to service the request, i.e., mostly allocation time. - Time until the chunk gets first used -- Less important, but good to get a sense of how much wiggle room there is in servicing requests. - Time to fill a chunk -- Interesting to get a sense of the rate of data generation. It could eventually inform an more intelligent request-service-handler, in case it would make sense for very busy processes to allocate more chunks in advance. Also, we should record whenever a chunk is allocated on the spot -- hopefully never!
Bug 1630094 Comment 0 Edit History
Note: The actual edited comment in the bug view page will always show the original commenter’s name and original timestamp.
Bug 1626837 introduces a new memory allocation scheme in the profiler (which will only start being used later on, see meta bug 1612356). Thanks to off-thread allocation handling, the overhead seemed low in some early manual testing, but it would be good to get more confidence (and also include potential overhead in profiles, so the user is aware of profiler-related overhead), and later fix issues if any. To be recorded: - Time between request of a new chunk, and when the request gets handled. - Time to service the request, i.e., mostly allocation time. - Time until the chunk gets first used -- Less important, but good to get a sense of how much wiggle room there is in servicing requests. - Time to fill a chunk -- Interesting to get a sense of the rate of data generation. It could eventually inform a more intelligent request-service-handler, in case it would make sense for very busy processes to allocate more chunks in advance. Also, we should record whenever a chunk is allocated on the spot -- hopefully never!