Re: GPU Process Experiment Results
Thanks for the feedback Chris. Graphs measuring crash rates is "crashes per 1000 hours" for Telemetry data via the crash_aggregates table (links to my re:dash queries are link at the bottom of the post) and "crashes per install_time cardinality" for Socorro data. I tried to focus on crash rate instead of crash count this time around. I'll endeavour to add more context in future graphs. Cheers! On 16 January 2017 at 13:22, Chris Hutten-Czapskiwrote: > My one-sentence summary of the article - If anything, the test cohort with > the GPU process saw improved stability, especially for graphics crashes. > > Which is awesome! > > May need error bars for the figures to see how much of the results you saw > we might have to attribute to the noise of reported crash figures. > > Also, I can't quite tell what the units are. The initial graphs seem to be > #crashes-reported-to-Socorro, but later ones talk of > #crashes-per-1000-usage-hours. The axes aren't labelled, so it's difficult > to be precise. > > Have you attempted to measure crash counts and types via Telemetry instead > of Socorro? If all you need is a count and some metadata, the submission > rate to Telemetry is much (25x) higher than Socorro. (actually, if all you > need is a count, crash_aggregates would be a good place to start, as most > of the counting has been done for you). > > All in all, very interesting and I can't wait to see future experiments in > Aurora and on Beta, when the time is right. > > :chutten > > > On Mon, Jan 16, 2017 at 4:11 PM, Anthony Hughes > wrote: > >> Hello Platform folks, >> >> Over the Christmas break I rolled out a Telemetry Experiment on Nightly to >> measure the stability impact of the GPU Process. This experiment concluded >> on January 11. Having had time to analyze the data I've published a report >> on my blog: >> https://ashughes.com/?p=374 >> >> It should come up on Planet shortly but I wanted to post here for >> increased >> visibility. Feel free to send me questions, comments, and feedback. >> >> Cheers >> >> -- >> Anthony Hughes >> Senior Quality Engineer >> Mozilla Corporation >> ___ >> dev-platform mailing list >> dev-platform@lists.mozilla.org >> https://lists.mozilla.org/listinfo/dev-platform >> > > -- Anthony Hughes Senior Quality Engineer Mozilla Corporation ___ dev-platform mailing list dev-platform@lists.mozilla.org https://lists.mozilla.org/listinfo/dev-platform
Re: GPU Process Experiment Results
My one-sentence summary of the article - If anything, the test cohort with the GPU process saw improved stability, especially for graphics crashes. Which is awesome! May need error bars for the figures to see how much of the results you saw we might have to attribute to the noise of reported crash figures. Also, I can't quite tell what the units are. The initial graphs seem to be #crashes-reported-to-Socorro, but later ones talk of #crashes-per-1000-usage-hours. The axes aren't labelled, so it's difficult to be precise. Have you attempted to measure crash counts and types via Telemetry instead of Socorro? If all you need is a count and some metadata, the submission rate to Telemetry is much (25x) higher than Socorro. (actually, if all you need is a count, crash_aggregates would be a good place to start, as most of the counting has been done for you). All in all, very interesting and I can't wait to see future experiments in Aurora and on Beta, when the time is right. :chutten On Mon, Jan 16, 2017 at 4:11 PM, Anthony Hugheswrote: > Hello Platform folks, > > Over the Christmas break I rolled out a Telemetry Experiment on Nightly to > measure the stability impact of the GPU Process. This experiment concluded > on January 11. Having had time to analyze the data I've published a report > on my blog: > https://ashughes.com/?p=374 > > It should come up on Planet shortly but I wanted to post here for increased > visibility. Feel free to send me questions, comments, and feedback. > > Cheers > > -- > Anthony Hughes > Senior Quality Engineer > Mozilla Corporation > ___ > dev-platform mailing list > dev-platform@lists.mozilla.org > https://lists.mozilla.org/listinfo/dev-platform > ___ dev-platform mailing list dev-platform@lists.mozilla.org https://lists.mozilla.org/listinfo/dev-platform
GPU Process Experiment Results
Hello Platform folks, Over the Christmas break I rolled out a Telemetry Experiment on Nightly to measure the stability impact of the GPU Process. This experiment concluded on January 11. Having had time to analyze the data I've published a report on my blog: https://ashughes.com/?p=374 It should come up on Planet shortly but I wanted to post here for increased visibility. Feel free to send me questions, comments, and feedback. Cheers -- Anthony Hughes Senior Quality Engineer Mozilla Corporation ___ dev-platform mailing list dev-platform@lists.mozilla.org https://lists.mozilla.org/listinfo/dev-platform