An IEEE1588 Precision Time Protocol Time Synchronization Algorithm implementation with a goal to minimize latency and network jitter.
-
Python: Probably the highest level implementation possible using standard sockets - Status: Done
-
C: Highest level implementation possible using C with standard socket system calls - Status: Done
-
C with kernel timestamping: One level lower using socket options to add kernel timestamping to reduce the non-deterministic delay from the time when the packet reaches the NIC to the time the userland records the time - Status: Icebox, too kernel/hardware specific, not enough support
-
C with kernel bypass using netmap: Bypassing the kernel completely using netmap and talking directly to the NIC using mmap, ioctl, and ring buffers. Should result in the most accurate measurements and fastest execution - Status: WIP
slave/ contains the implementation for the slave, which is a server that will take requests to sync from the master.
master/ contains the implementation for the master which will initiate the sync protocol with the slave.
Run the slave on the designated slave machine first, then run the master from the designated master.
Make sure you configure the slave IP address and the number of times to run the protocol at the top of libs/common.h. This implementation assumes that the master is a 64-bit machine, but the slave can be 32-bit.
For instructions on building, run "make" in the root directory. The makefiles in the individual server and pi directories are for debugging purposes.
Netmap: http://info.iet.unipi.it/~luigi/netmap/
IEEE1588 explanation: http://www.nist.gov/el/isd/ieee/upload/tutorial-basic.pdf