Questions for 6.S897 lecture 6 (Storage 2, 09/29).
Email your answers to 6.s897staff@gmail.com.

1) In the BigTable paper, what was the network bandwidth used per node when all 500 nodes were doing random reads, in figure 6? What about when all of them were doing random writes? (To answer this, also discuss how much network bandwidth one random write should use).

2) The Dynamo paper mentions that most of the reads that required reconciliation (i.e. returned more than one version) happened due to concurrent writes. How can concurrent writes lead to multiple versions in Dynamo?

3) Compare BigTable, Dynamo and TAO along the following three dimensions: data model, consistency guarantees, and availability.