Consistent time service for fault-tolerant distributed systems

A method of maintaining clock consistency for a group of replicas having physical hardware clocks and operating in a fault-tolerant distributed system, comprising:

• reading a physical clock value for a given replica in response to a clock operation for a thread;
• determining a local logical clock value for said given replica by adding a clock offset value to said physical clock value;
• proposing said local logical clock value as a group clock value within a clock synchronization message sent to replicas in said group of replicas when no other clock synchronization messages for the calling thread for this reading of the physical clock value have been received;
• extracting said local logical clock value from a received clock synchronization message which is established as a group clock value that is consistent across said group of replicas;
• updating said clock offset value for said given replica to the value of the group clock value less the physical clock value;
• and
• returning said group clock value to said thread of said given replica.

A method of maintaining clock consistency for a group of replicas operating in a fault-tolerant distributed system, said clock consistency maintained as said replicas access their respective physical hardware clocks through a time service handler, comprising:

• processing clock synchronization messages that arrive at a given replica from other replicas within said group of replicas;
• updating a time offset value for said given replica as the difference between the physical hardware clock value for said given replica and a clock value contained in said clock synchronization message;
• determining a local logical clock value from the sum of said physical hardware clock value and said time offset value;
• proposing said local logical clock value by said given replica as a group clock value for replicas in said group of replicas by multicasting a clock synchronization message to the replicas in said group of replicas;
• establishing said proposed clock value as a group clock value by other replicas in said group of replicas;
• updating said clock offset value by each said replica accepting said group clock value;
• and
• returning said group clock value to said given replica from said time service handler.