Bigtable: A Distributed Storage System for Structured Data
-
How does master discover the tablet server? How does client discover the tablet server? Compare and contrast the solution with similar problems in GFS
-
How does bigtable detect master health? What will happen if IT THINKS the master has failed? Compare this failure detection and recovery with GFS’s design on its master and primary chuck server
-
How does master detects tablet server’s health? What happens when the master thinks the tablet server has failed? Compare this interaction with GFS’s between its master and chunk server
-
How can master learn the the newly split tablet? How does GFS master learn the newly created chuck?
-
The commit log is in GFS. What did they do in case GFS is experiencing latency? How would recovery work in this case?
-
SSTables are in GFS already, why do we still need that commit log?
-
Sequence of actions on read? compare this with GFS’s sequence of reading
-
Sequence of actions on write? compare this with GFS’s sequence of writing
-
What kinds of info/metrics they monitor, and how did they do that? Compare their collections with GFS’s.