The Network Time Protocol (NTP) is widely used to synchronize computer clocks throughout the Internet. Existing NTP clients and servers form a very large distributed system, and yet the tools available to observe and manage this system are fairly primitive. This paper describes our experiences with a prototype tool that attempts to discover relevant information about every NTP site on the Internet. The data produced by this tool can be used for a variety of purposes, including locating nearby accurate time servers and computing aggregate and long-term evaluations of the size and health of the NTP system. Importantly, our tool provides a means by which new NTP server administrators can make informed choices among the possible servers with which to synchronize, balancing the need for accurate time with the need to distribute server load. This is an important step towards improving global NTP system scalability, since at present our measurements indicate that the high-stratum servers are...
James D. Guyton, Michael F. Schwartz