*Question*: Is it "better" in Chromium to use the HIGH_MEMORY model, or the MEDIUM_MEMORY model (re: command line flag "memory-model")? Our metric for goodness is the tab-switching time, especially when a user has been away from the application for an extended period of time. *Answer*: Since we weren't sure which was "better" (default is MEDIUM_MEMORY), I landed some code to do a field trial and (try to) find out.
*Who is in this field trial experiment?* Note this code is not yet pushed to either Google Chrome's dev channel, nor beta channel. I've only landed this on the Chromium tip-of-tree thus far. It is possible that we'll get enough good data that we won't even need to run the trial on the beta channel. If you don't set a command line switch (demanding a preferred model), the code will "put you in the experiment." Runs in the experiment will randomly select between the two memory models each time Chromium starts up. In addition, the code will histogram tab-switch time into a slightly different histogram depending on which group the process is running in (randomly selected HIGH vs MEDIUM). As a result, I'm hopeful that we'll end up with two distinct histograms to compare, and get some insight into the impact of the choice on a real user population. If the user opted-in to helping us improve Google Chrome, the distinct histogram data that is gathered will be uploaded and aggregated for analysis. As with any such data, no personally identifiable info will be uploaded, just time-deltas for switching between tabs, with no record of the contents etc. of the tabs. *How do I avoid this experimental randomization of my runs?* Manually select a memory model on the command line. I'm sending out this note mostly to warn of the minimally non-deterministic behavior I've added. If you don't want to "be in the experiment," just use a command line flag to demand the memory model of your choice. *Will this impact the build-bot perf tests?* I suspect that the value will have little impact on the bulid buts, as it relates to swapping out of RAM, which will probably not be an issues for short-lived build-bot tests. If I cause too much chaos (variance) on the bulid-bot perf tests, we may perhaps need to nail this command line value down. ------------------------------------------------------------------------------- Comments and suggestions welcome. Note that I'm currently only splitting out the whiteout-duration histogram (used to measure time when tab switching), as the hope is that this command line flag will impact tab-switching responsiveness. Mbelshe also suggested we provide, for users in the experiment, distinct memory use histograms, to see if those values are also impacted. Thanks, Jim --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "Chromium-dev" group. To post to this group, send email to [email protected] To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/chromium-dev?hl=en -~----------~----~----~----~------~----~------~--~---
