Method and apparatus for lockstep processing on a fixed-latency interconnect

A method comprising:

• receiving a message from a node on a communication network;
• allocating a buffer to hold an acknowledgement;
• sending the message to one of a plurality of devices;
• receiving an acknowledgement from the one of the plurality of devices;
• and
• processing the message.

A method comprising:

• receiving a message from one of a plurality of devices;
• sending an acknowledgement to the one of the plurality of devices over a known-latency communication medium;
• waiting a predetermined period of time;
• and
• processing the message.

A processor comprising:

• an interconnection network;
• a plurality of processor cores connected to the network, wherein a first core is to operate in lockstep with a second core;
• a device connected to the network;
• mapping logic to determine relative positions of the first core, the second core and the device on the network and to identify which of the first and second cores is closer to the device and which is farther from the device;
• request logic to coordinate requests from the first and second cores to the device;
• and
• response logic to coordinate responses from the device to the first and second cores.

A system comprising:

• a plurality of central processing units (“CPUs”);
• a memory;
• and
• a network to carry data between the CPUs and the memory;
• wherein
• a message from the memory is to arrive at a first CPU through the network;
• the first CPU is to send the message to a second CPUs through the network;
• the second CPU is to send an acknowledgement to the first CPU through the network;
• and
• the second CPU is to begin processing the message a predetermined period of time after sending the acknowledgement.