dknopik week16

[dknopik]

It's pretty long, sorry for making you read all that, Mario ;(

Some links don't work currently, as ProbeLab's site is down, hopefully they are back soon

[taxmeifyoucan]

Thanks for the update Daniel! Definitely worth the read :) Lot's of interesting data, makes me really curious for more simulations. I believe the report will make an important contribution to current discussions around peerdas. But I think even if you make it before end of the cohort, more simulation might be needed as time goes on. Contributing to shadow, creating and documenting the reusable infrastructure is a great approach.
One thing I didn't find in the update - did you consult the data with any mentor? I wonder what e.g. Francesco or people working on the spec would think about tweaking the parameters.

[dknopik]

@taxmeifyoucan
I kept Joao, Jimmy and Age from Sigp in the loop during the simulations and provided and discussed results with them. Some of the experiments were not thought out yet when the first ones started as I needed to figure out the issue first. My discussions with them helped a lot with finding the next steps and confirming my findings!
As briefly mentioned during the call just now, I don't think the data is viable to decide anything for the spec yet. I feel like as soon as Lighthouse properly implements at least two of a) better peer selection, b) data column broadcast reattempts or c) some kind of fault-tolerant sampling, someone should run some of these experiments again. :)