README driven development! Here we go!
Technically there is no reason not to run MapReduce jobs on idling web clients. But of course this was a competition and we had to spend some time to implement cool realtime scrolling graphs, so some things were left out.
- There is no input reader step. Data has to be uploaded in chunks.
- There's always exactly one map phase and one reduce phase. No partitions and you can't supply a partitioning function.
- The grouping step after the map step is lame and slow.
- Results aren't validated by more than one client.
- We're using a relational database goes and hides in the basement
- This blog post: http://www.igvita.com/2009/03/03/collaborative-map-reduce-in-the-browser/
- My mother! My father! My sister and my brother!