dpowles@CCD.BBN.COM ("Drew M. Powles") (05/11/88)
Can someone please tell me which protocol is the most widely accepted for synchronizing Network Time? What work is being done in this area? What do most vendors implement? Is it RFC 958? Is there a later version? Thanks in advance for any and all information. Drew Powles BBN Communications dpowles@bbn.com
Mills@UDEL.EDU (05/12/88)
Drew, First, let me point out that I am far from an unbiased source. Probably the most widely implemented time protocol is UDP/TIME, which is distributed with vanilla Unix. The 4.3bsd Unix distributions include a time daemon which manages time distribution within a LAN, but would not be suitable for use over most multi-net Internet paths. The Network Time Protocol (NTP) described in RFC-958, as amended, was specifically intended for time distribution both within a LAN and as far as the Internet eye can see. A 4.3bsd daemon for NTP is available from MIke Petry (petry@trantor.umd.edu). There are varying opinions on the robustness and accuracy of various time-synchronization mechanisms in the literature. The NTP design was based on models used by digital common carriers and uses maximum-likelihood estimation and nonlinear-filtering techniques. A survey of different approaches to the problem, the factors the led to the NTP design choices and a comprehensive bibliography on the subject can be found in the ~ftp/pub/ntp.doc file in the anonymous directory at louie.udel.edu. This document has been printed as a departmental report and submitted as an RFC. In my opinion it does not make sense to porpose a new service without extensive experimentation and evaluation of a prototype implementation. NTP was born three years ago and has been in regular use since that time with at least three radio-synchronized time servers (now with five). As the result of this experience the original NTP design was thoroughly overhauled and tested over the last six months. The resulting NTP design, including algorithms necessary to improve accuracy and mitigate between possibly broken or bogus clocks (falsetickers), is described in the cited document. All five of the radio-synchronized NTP servers, as well as all Fuzzball secondary servers (e.g. NSFNET Backbone gateways) and many Unix secondary/retail servers now tick the new version. In most places where such things can be measured, the service provides time acccurate to within a few tens of milliseconds and has an incredible degree of redundancy. Dave P.S. Truechimers may wish to subscribe to the list ntp@trantor.umd.edu for additional information. DLM