When Bully and Ring algorithms suit

A) Bully algorithm is suitable when

• The system is small to medium and all processes know the ids of all others and can communicate directly (fully connected or a reliable network).
• Fast recovery is wanted: when the failure is detected by a high-id process, the election completes quickly with few messages.
• The system has reliable, bounded-delay (synchronous) communication, so a timeout reliably indicates a crashed process.
• The highest priority/id process (the most powerful or reliable machine) should always be coordinator; a recovering strong node should take over (it "bullies" its way in).
• A message overhead of O(N²) in the worst case is acceptable. Examples: a LAN cluster with a primary database server, replicated file servers.

B) Ring algorithm is suitable when

• Processes are already organised in a logical ring (for example, token ring systems), and each knows only its successor (the full membership list is not known).
• The number of messages should be predictable and bounded (about 2N) regardless of who starts, or many processes may start elections simultaneously (all get the same result).
• The network is not fully connected, or point-to-point links to all processes are expensive.
• Slower elections are acceptable, since the message goes around the full ring, and the ring can be repaired by skipping failed nodes. Examples: token ring networks, ring-structured peer-to-peer overlays, and token regeneration in a token-ring mutual exclusion system.