Skip to end of metadata
Go to start of metadata


Andrey, Peter, Arnau, Alex



  • the input stream is sorted by timestamp now
  • read queries should be less frequent than update queries. Experiments are needed to figure out the specific proportion
  • the strategy for getting the parameters is as follows: we mine the static dataset (used for bulk loading) to get the parameters, and ignore the dynamic one for now. The assumption is that the time distribution for those parameters will stay the same when executed on the entire dataset. Possible exception is the tag counts (due to the bursty behavior) 
  • No labels