We do have a contradiction here, if that's what you're saying.
I was not in YT. I was only told their backend was Python by people who were. I can't explain that paper OP saw. Could be it reflected reality and I'm wrong; could be it was only a design.
I do know for a fact that they still had their own experiment infrastructure, which does suggest that the backend was still legacy.
I was also not in YT (was in Search from 09-14), but a bunch of my coworkers went over to work there around '10-11. I read the paper before joining Google; I think it came out in 07 or 08.
There might also be a lot of confusion over "backend" vs. "frontend" terminology. Like the other poster mentioned, "frontend" at Google usually means the webserver and sometimes extends back to various application-level services. Basically everything that involves user interaction. Experiment infrastructure has been part of the frontend in every system I've worked with at Google (which now includes Search, Google+, GFiber, Doodles, AndroidTV, and Assistant), though the flags often get plumbed back to backends to alter behavior there.
By "backend" I mean the storage & offline processing areas. BigTable, Colossus, MapReduce, various blob stores, training machine-learning models, etc. This is the part of YT that (in my understanding) was rewritten to use Google technologies.
iamstupidsimple|4 years ago
AlbertCory|4 years ago
I was not in YT. I was only told their backend was Python by people who were. I can't explain that paper OP saw. Could be it reflected reality and I'm wrong; could be it was only a design.
I do know for a fact that they still had their own experiment infrastructure, which does suggest that the backend was still legacy.
nostrademons|4 years ago
There might also be a lot of confusion over "backend" vs. "frontend" terminology. Like the other poster mentioned, "frontend" at Google usually means the webserver and sometimes extends back to various application-level services. Basically everything that involves user interaction. Experiment infrastructure has been part of the frontend in every system I've worked with at Google (which now includes Search, Google+, GFiber, Doodles, AndroidTV, and Assistant), though the flags often get plumbed back to backends to alter behavior there.
By "backend" I mean the storage & offline processing areas. BigTable, Colossus, MapReduce, various blob stores, training machine-learning models, etc. This is the part of YT that (in my understanding) was rewritten to use Google technologies.