Electronic Seismologist logo

ELECTRONIC SEISMOLOGIST

September/October 2003

Thomas J. Owens
E-mail: owens@sc.edu
Department of Geological Sciences
University of South Carolina
Columbia, SC 29208
Phone: +1-803-777-4530
Fax: +1-803-777-0906

It's All about Time

Last fall, the ES faced a dilemma. He could spend time writing a full-length article for the special SRL Educational Seismology issue by late September OR wait nine months and write an ES column for the issue. Guess which choice came out on top? Yup, the ES procrastinated. Then, as the June deadline approached, a third choice presented itself: Have someone else write the column! So, in the spirit of the Educational Seismology issue, our lead guest author this month is Andrew Frassetto, a rising senior Geophysics major at the ES's home institution. Andrew has been working with the ES and his famous sidekick, Philip Crotwell, on the South Carolina Earth Physics Project (SCEPP) for a couple of years, getting a true education in both the challenges of modern seismology and the challenges of working in the K-12 educational system.

For the ES, working on the SCEPP project has provided some of the most satisfying and most frustrating experiences of his professional career. Teachers are some of the most interested and motivated students that the ES has ever encountered. However, they and everyone in their school districts are very busy, which caused the ES more than minor frustration when trying to get their attention to install a seismograph station in their school. The ES and other SCEPP staffers quickly learned that minimizing the efforts that teachers and school district personnel need to make on your behalf is the single most important factor in getting instruments into schools. Having schools prepare outdoor sites was extremely difficult; asking them to locate instruments near outside walls where we could have GPS signals and Internet connections was merely difficult. Fortunately, asking them to find a location anywhere they wanted within reach of an Internet drop was something most schools could handle in the limited time they had available. But does saving a teacher's time compromise the time accuracy of the seismograph station, a sacred issue among seismologists? When we adopted the Network Time Protocol (NTP) for the SCEPP network, we didn't really know the answer to this question. We just knew that it was a solution that made life easier for our teachers. Through Andrew Frassetto's efforts, we can now say that the timing accuracy using NTP is quite acceptable for educational seismology applications and likely tolerable for many research applications. Thanks, Andrew!

Evaluating the Network Time Protocol (NTP) for Timing in the South Carolina Earth Physics Project (SCEPP)

Andrew Frassetto, Thomas J. Owens, and Philip Crotwell
Department of Geological Sciences
University of South Carolina
Columbia, SC 29208
Telephone: +1-803-777-8986
Fax: +1-803-777-0906
E-mail: andyf@seis.sc.edu

First Author Commentary: In January 2001, Tom Owens recruited me as an undergraduate assistant with the South Carolina Earth Physics Project. As a novice to the field of geophysics, I welcomed the opportunity to augment my limited knowledge on the subject while also earning an income without having to bag groceries or wait tables. Little did I realize that my few hours work per week as network manager would expand into one of the most educational experiences in my undergraduate career. When I signed onto SCEPP, I lacked a solid foundation in the principles of modern seismic data acquisition. I had no concept of the methods, software, and instrumentation that functioned to create the SCEPP network. When Tom and Philip first introduced me to this array of foreign-looking equipment, I realized that modern seismology fully utilizes the digital revolution. At SCEPP, there are no pen and ink drum recorders. Nearly every step from data recording to data processing and visualization occurs in the digital realm. We collect data using PMD seismometers connected to a Symmetric Research digitizer and a Linux-powered Pentium computer. These computers apply the time at which the signal is recorded and relay the data to our server at USC. My research task sounded simple at the time: evaluate SCEPP timing.

In seismology, station clock accuracy is essential in producing quality seismograms with research potential. To address the need for accurate clocks, but do so with budget limitations, SCEPP uses Internet-based time signals rather than locally installed GPS-based time signals to maintain software and hardware clock accuracy in our network computers. In particular, SCEPP utilizes the Network Time Protocol (NTP; http://www.ntp.org/). Since the software and hardware clocks of a computer are subject to continuous drift of varying degrees, the inclusion of NTP software in our seismic stations allows us to keep accurate time during the recording process. NTP functions by accessing specified NTP servers that use GPS clocks to maintain time with a high degree of accuracy. Thus, it saves the time and expense of installing GPS receivers on each school seismograph site. With the use of NTP, SCEPP seismographs are able to keep good clock accuracy by continuously verifying their local time with the time recorded by the NTP servers. The only factor that can inhibit timing accuracy pertains to Internet issues that may lengthen the travel time between NTP servers and seismic stations on some occasions. Almost immediately after the primary deployment of SCEPP stations across South Carolina, we began pondering NTP's accuracy. NTP times are reported across our Internet connection from five separate servers that are all in the vicinity of the eastern U.S. (Figure 1). NTP servers are assigned stratum numbers based on their distance from an accurate time source (primarily GPS clocks, which are assigned Stratum 0). An NTP client can lock to any NTP server with a stratum number less than its own. The NTP software monitors transmissions to/from servers designated by the user and determines which server is "best" for clock synchronization. Numerous papers have been published about NTP over the last decade (e.g., Mills, 1990) and it is supported by a labyrinthine help site (http://www.ntp.org/). We won't try to duplicate this body of literature in any way. The efforts in these papers primarily focus on establishing the statistical basis for understanding the accuracy of time obtained via NTP given the unpredictable variability of transit times of the Internet. The unsettling part of using NTP is that the time errors are only estimates. The good news is that these estimates are supported by many years of analysis by the NTP community. And, if all that is not enough, the following analysis by a geophysics undergraduate should be enough to convince you to trust NTP!

Figure 1

Figure 1. Map of active (triangles) and defunct (circles) SCEPP stations as of June 2003. Squares indicate locations of NTP servers at the University of South Carolina (Columbia) and the University of Georgia System (Atlanta) that are used by SCEPP stations for time synchronization. Additional NTP servers at Penn State University, University of Illinois, and University of Delaware are available to SCEPP stations if necessary. Stations used in Figure 2 are indicated by name.

 

Figure 2

Figure 2. Clock lock offsets for ten SCEPP stations for a three-day period in May 2002. Vertical scale in the center of plot is used for all stations. Locations of stations analyzed are shown in Figure 1.

 

NTP continually attempts to verify the local computer's time with designated servers. When this continual polling indicates a stable link between the local machine and a remote server, NTP locks the local clock to the remote server and records an estimated offset, which can be saved in log files. A study of NTP function and accuracy by Mills (1990) explained that clock offset is determined by summing the difference between remote and local server times and the difference between the local and remote server on the return trip. There is obviously more to it than this, as the current generation of NTP software does considerable analysis of the exchanged data to improve the estimates. Clock offset thus reflects the difference between the actual GPS corrected time at the NTP reference server and the time to be adjusted at the SCEPP computer. An additional feature of NTP is that it monitors the drift of your local machine's clock and can correct for this drift when your machine is not locked to an NTP server. Over time, theoretically, this should further improve the time accuracy of your local machine.

All SCEPP stations at schools record and archive the NTP logging. We do not yet correct the recorded seismograms for the offsets indicated by NTP. Widespread use of NTP in educational seismology could make the development of correction software worthwhile. We chose a three-day period in early May 2002 to illustrate the performance of NTP within the SCEPP network. Data from the NTP log files were extracted for the nearly 17,000 declared clock locks at ten SCEPP stations. Most stations prefer to lock to one server most of the time. When a long-term server switch occurred this was often preceded by a period of "instability" in the recorded offsets, possibly indicating heavy network traffic or a change in topology that triggers packet rerouting. Overall, recorded clock offsets are small (Figure 2), almost always less than 20 msec (Figure 3). Offsets greater than 0.1 seconds were detected only nine times in the three-day test period, indicating that NTP provides a stable time base for our recording systems.

FIgure 3

Figure 3. Histogram of clock lock offsets for all ten stations for the three-day period analyzed in Figure 2. Nearly 17,000 clock locks occurred, with offsets closely clustered around zero. Nine offsets occurred outside the limits of this plot.

 

All the statistical analysis in the NTP literature caused the eyes of all authors of this note to glaze over on occasion. So we devised our own test to evaluate NTP errors produced by network traffic in a more understandable context. We configured one machine at USC as a primary NTP server and then set up a test client machine right next to it in our lab and a SCEPP station (CLOVE) to lock only to that server. CLOVE was chosen because it is one of the more remote stations relative to USC (and because it had a broken sensor at the time!). Both CLOVE and our test client were configured to record offsets relative to each other as well as relative to the primary server. The test client machine in our lab would have essentially zero network delays relative to the server sitting next to it. Thus, we could analyze offset circuits between the server and the two clients and verify the consistency of the NTP log files. A direct comparison was not possible because the times of recorded offsets were not synchronized between the local and remote clients. When the offsets were recorded at nearly identical times, however, the results were consistent within a few milliseconds. This gives us some confidence that the offsets recorded by the NTP logging mechanism are reflecting real offsets relative to our designated servers.

More detailed studies could be done to build a deeper understanding of NTP errors if we wanted to be able to apply specific corrections to our data for, say, precise time-picking. However, SCEPP is an educational network and the site noise is high as well. Thus, we believe that the 20 msec or so accuracy of NTP is excellent for the intended purpose of SCEPP given the advantages that it provides. Success in educational seismology depends on removing obstructions to a teacher beginning to participate in these activities. There is no doubt that direct GPS timing on a seismograph station can provide better data for correcting that station's timing. However, for a teacher with a laboratory or preparation room without any outside walls, the additional requirement of an appropriate location for a GPS antenna could be enough of a barrier for that person to decide not to participate. Ease of access and operation, both physically and intellectually, has been a goal of every aspect of SCEPP. Use of the NTP software allows us to maintain a relatively simple hardware set-up and reduces the need for hardware maintenance. Most, if not all, PC's now come with NTP already installed. If an educational seismograph kit came with instructions on how to check/enable NTP, then a teacher would be free to put the instrument anywhere and perhaps more students would begin to benefit from learning with earthquakes.

References

Mills, D. L. (1990). On the accuracy and stability of clocks synchronized by the Network Time Protocol in the Internet system, ACM Computer Communication Review 20, 65-75.


SRL encourages guest columnists to contribute to the "Electronic Seismologist." Please contact Tom Owens with your ideas. His e-mail address is owens@sc.edu.

 

HOME

Posted: 21 June 2005