PracticeQ Replication

1. What is the benefit of a state-transfer based approach compared to an RSM in a replicated system?

It can capture non-deterministic operations

2. Alice wishes to replicate her board game server. A core operation in her system is save_dice_roll(“key”, number_of_sides_of_dice). Using an RSM is a good replication strategy for Alice:

False, because the roll of a dice is non-deterministic, operforming the same operation on another server may have inconsistent result

3. When does the “commit point” occur in chain replication?

e. When the tail server responds to the client

4. We make the following modification to primary-backup replication: the primary may respond to a client write, after only after one backup has acknowledged the update (instead of all backups). Explain how, with this plan, a client may have a stale read occur?

If primary crashes, CFG delegates a server, which had not yet received the propagation as the primary, the client reads a not up-to-date value

5. In chain replication, the configuration server (CFG)

b) is the source of truth for which servers are live in the chain

6. In primary-backup replication, when a client performs a read, we must communicate with the backups.

No, because all servers have up-to-date value, and also primary is responsible for all client requests

7. If we have a workload of 100% reads, which method of replication has better throughput?

Same. Both chain replication and primary backup replication are bottlenecked by a single server in the case of 100% reads. In chain replication with only reads, 100% of the requests will go to the tail server, in p/b,100% of requests go to the primary.