Both GFS and Bigtable make the same core design choice – to have a single master. What are the repercussions of a failure of this single master in each case?

What will be an ideal response?

As mentioned above, the Google philosophy is to provide minimal services that do one task efficiently. Google therefore provide three minimal services all doing their own job well, which together provide coverage of their storage needs:
• GFS offers storage of very large datasets optimized for the styles of reads and writes encountered in their environment;
• Chubby offers efficient access to very small data items, together with the services for distributed consensus;
• Bigtable supports access to semi-structured data (building on top of GFS and Chubby). Google would not want to use a commercial distributed database for three reasons:
• they prefer to own and be responsible for their entire code base;

• distributed databases are not tailored towards their own operating environment, for example dealing with massive datasets;
• distributed databases, for example relational databases, offered rich functionality that goes beyond what is necessary in the Google environment (for example, full relational operators) and this also comes at a price in terms of performance.