home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.protocols.time.ntp
- Path: sparky!uunet!europa.asd.contel.com!darwin.sura.net!wupost!gumby!destroyer!ubc-cs!unixg.ubc.ca!kakwa.ucs.ualberta.ca!access.usask.ca!herald.usask.ca!glenn
- From: glenn@herald.usask.ca (Glenn Hollinger)
- Subject: NTP and DTSS at the University of Saskatchewan
- Message-ID: <1992Sep9.174049.24379@access.usask.ca>
- Originator: glenn@herald.usask.ca
- Sender: news@access.usask.ca (USENET News System)
- Nntp-Posting-Host: herald.usask.ca
- Organization: University of Saskatchewan
- Date: Wed, 9 Sep 1992 17:40:49 GMT
- Lines: 121
-
- We are now using DEC DTSS and NTP to synchronize our campus computers.
- This seems to be a recurring topic of questions, so I put together some
- information on what we have done here. First, an ascii diagram of the
- layout of things here and then a rather lengthy history of how things
- progressed. As with all things, discussion is encouraged. I will
- answer questions if I can and post a summary to the net.
-
- Glenn Hollinger glenn@herald.usask.ca
- University of Saskatchewan +1-306-966-4827
-
-
- Physical layout:
- _______Ethernet_____________________
- | | |
- GOES MicroVAX 2000 & UNIX VMS
- satellite ---RS232---------> VAXstation 2000 Systems Systems
- clock
-
- Time synchronization flow:
-
- GOES Ascii Xntp3 synchronized
- satellite ---Time----> daemon ---adjtime calls----> local
- clock Codes (Ultrix) clock
- / \
- / \
- V V
- V3 V1, V2, V3 NTP
- Broadcast NTP Peer-to-Peer
- / | \
- / | \
- V V V
- V3 NTP NTP peers DTSS NTP
- broadcast clients (Unix) Time Providers--->DTSS Server
- (Unix) (VMS) (VMS)
- |
- |
- V
- DTSS Clients
- (VMS)
-
-
- Traveling Time Using Xntp and DECdts
- or
- How to sink 150 computers without really trying
-
- This text is a chronological history of time synchronization at the University
- of Saskatchewan. Like most campuses, we have many independent computer
- systems. Each of these systems had their time set at install time using the
- installer's wrist watch as the accurate time source. After that, they ran
- independently and incensed.
-
- Some operating systems do use various methods to synchronize, such as the rdate
- techniques of Ultrix and SunOS (which are different from each other). This
- helped those computers somewhat, but didn't solve the problem completely, even
- for them.
-
- About September of 1990 the brand new CA*net network was installed with its
- high-speed 56 kbit backbone spanning the country and linking our campus to the
- internet. At this time, we saw the opportunity to use the NTP servers on the
- internet. By December of 1990, we started using NTP to synchronize some of our
- Unix computers. Not all OSes could be handled, but this was a start.
-
- Using NTP version 1, we synced up most of the campus Ultrix and SunOS boxes
- (and other unixes). This worked fairly well getting time from Stratum 1 NTP
- servers in the states. But, CA*net was such a success that it started getting
- busier, especially at the CA*net-Internet links. Even with line speed
- increases in critical areas, by December 1991 decent synchronization to US
- servers was becoming harder to get.
-
- We needed a plan. The plan we came up with was to make a decent local clock we
- could rely on. In February 1992, we scrounged a MicroVAX 2000 running Ultrix.
- This node with Xntp software made a decent stratum 2 NTP server, but didn't
- solve our troubles communicating to US stratum 1 servers. We started casting
- about looking for accurate time sources.
-
- The first thing we latched on to was CHU. This turned out to be a dead end
- because of difficulties establishing reliable radio reception.
-
- Next, we looked into a satellite clock our Geologists were using, a Kinimetrics
- Truetime 468-DC. This is a nice time source, but not supported by any NTP
- software at that time.
-
- From February through May 1992 we took xntp3 beta and developed a GOES clock
- module to support the GOES satellite clock. Several iterations were done with
- long-term testing for problems throughout. The final result is a Stratum 1 NTP
- server on our local ethernet.
-
- Using broadcast NTP as well as NTP versions 1 2 and 3, it was relatively
- straight forward to distribute this time to Unix systems throughout the campus.
- However, this did not solve the problem for our 75 or so VMS systems.
-
- We heard that DECdts might solve this problem. After looking into it, we found
- out that Decnet 5.0 for Ultrix includes the DTSS server and could pick up time
- from an NTP server or a local ntp-synced clock. Later, we found out the DECdts
- in the VMS Decnet extensions can also talk to remote NTP servers if a public
- domain extension was used.
-
- After 2 months work, we discovered that the only way to use NTP and dtss on the
- same node is to prevent NTP from adjusting the system clock and let dtss do it.
- We felt this was an unreasonable limitation. So we junked OSI and, using
- public domain extensions to the VMS DTSS time provider, set up our DTSS servers
- on VMS to pick time up via UCX from the NTP servers. These DTSS servers then
- synchronize the VMS DTSS clients on the network.
-
- As MicroVax and VAXstation 2000s are becoming less useful, we have picked up a
- second system to be a redundant NTP server and a third system for spare parts.
- With a combination of scrounged hardware, some software development and some
- vendor supplied software, we will soon have our campus systems synchronized
- within about one hundredth of a second.
-
- Future directions; The GOES clock is not the best timekeeping device. We will
- be constructing the gadget box as described in the xntp3 distribution to
- connect its 1PPS output. This will hopeful improve accuracy quite a bit.
- Also, the GOES clock is not ours. We are looking at the new Motorola GPS
- receiver as a replacement which can provide a 1PPS signal with 50 nanoseconds
- accuracy for a reasonable cost.
-
- Glenn Hollinger,
- Systems Group,
- Computing Services,
- University of Saskatchewan.
-