M9.0 Sumatra Andaman Island Earthquake, December 26, 2004

The M9.0 earthquake centered in the eastern Indian Ocean has been widely reported by news media around the world. This has created a huge surge of traffic on the earthquake program web servers. Here is a quick run-down of the characteristics of this surge.

Figure 1 Figure 2
Figure 3 Figure 4
Figure 5 Figure 6
Figure 7 Figure 8
Figure 9

Figure 1 shows the traffic on the Earthquake Hazards Program web site for the month of December. This is from the statistics for the site, made from the logs from Akamai. The traffic surge is clearly visible on the graph, starting on December 26th. This surge is qualitatively different from surges generated by earthquakes that are directly felt by our readers. In the case of a felt earthquake, traffic surges quickly to a sharp peak, usually within five minutes of the event. The traffic surge generated by this event has been driven by news reporting, so it has been more gradual, but also much more long-lasting. Note that for most of the month, the daily average is about 6-8Gbytes/day served, but this jumps up to 153GBytes on the 26th, which is the first day after the earthquake. This works out to an average of about 14 Mbits/sec for the 24-hour period. Also note that as of this writing, about 65% of the month's traffic has been in the three days since the earthquake.

Figures 2, 3, and 4 show the same information for the NEIC, Menlo Park, and Pasadena web sites. All the graphs look similar. The peak 24-hour bandwidth usages for the sites are:
Menlo Park15.2GB

Note that the peak bandwidth usage on the main earthquake site occurred on the 26th, while the peak bandwidth on the other sites occurred on the 27th. This suggests that we may have been successful in our quest to teach the public to go to 'earthquake.usgs.gov first.

Figure 5 shows traffic on the Earthquake Hazards site for the 24 hours beginning at 00:00GMT on December 26th, just about one hour before the earthquake. The event occurred at 00:58GMT on the 26th. Note that traffic grew almost linearly for about 16 hours.

Figures 6, 7, and 8 are the corresponding graphs for NEIC, Menlo Park, and Pasadena. All of them show growth throughout the day, but only the NEIC graph duplicates the shape of the Program site graph. This is most likely because most of the specific information about this event is served off of the NEIC, and the 'Recent Earthquakes' maps on the Program site all link to the NEIC site.

Akamai reported that total bandwidth for all the USGS earthquake sites peaked at about 60Mbits/sec on that day, with an average total of about 30Mbits/sec. So in this case, the main Earthquake Hazards site is serving about half the total traffic, with the other sites (Menlo Park, NEIC, Pasadena, Utah, usgs.gov) making up the other half. This illustrates that 'a rising tide lifts all boats' in that reporting on a large earthquake drives traffic to all earthquake web sites, regardless of whether they have information pertaining to the particular event. Also, the Pasadena site is still the home of CIIM, which has gone global, so the Pasadena site is serving the 'Did you feel it?' map for this event, as well as acting as the front-end for collecing the online questionnaires for the event.

The Earthquake Hazards web site is served through the Akamai EdgeSuite service, so it has not experienced any availability problems after this event. The Akamai service acts as an amplifier on our web servers. The bandwidth for the Earthquake Hazards servers can be seen in Figure 9. This graph shows six days on the server in Menlo Park. The servers in Denver and Reston show almost identical graphs. The event occurred at 00:58GMT on the 26th, which was 16:58 in California where this server is located. The peak bandwidth at recorded by this server is about 1.6Mbits/sec. Multiplying this by 3 for the three servers puts our bandwidth usage at about 5Mbits/sec, which shows that the Akamai service is giving us about an order of magnitude amplification of our web server capacity.

The peak bandwidth usage on the back-end servers for Menlo Park and Pasadena was about 1.5Mbits/sec for Menlo Park and about 1Mbit/sec for Pasadena. I don't have access to this information for the NEIC site, but using the front-end/back-end bandwidth ratio for the other sites allows for a reasonable estimate. Most likely, the peak back-end bandwidth usage at NEIC was about 15-20Mbits/sec. I don't know what kind of hardware the NEIC server is running on these days, but even the very powerful Program servers would be feeling some strain at that level of traffic. This is likely related to some reports of files on the NEIC site being slow to load. Also, it is possibly related to observations from people in Golden that their connection to the Internet through Denver has been slow.

Overall, the Earthquake Hazards web sites have been performing well through the surge of traffic generated by this event.

Stan Schwarz
Honeywell Technical Services
Southern California Seismic Network Contract
Pasadena, California
December 29, 2004