M4.3 and M4.1, January 13, 2001: Web Service Report
A pair of events, M4.3 and M4.1 occurred on the evening of Saturday, January
13, 2001. These events were centered in the northeast San Fernando Valley
area of Los Angeles. They were felt by many people, and provided a good
real test of the USGS Pasadena Office and Trinet web service capabilities.
The Internet is fast becoming the medium of choice for timely
distribution of earthquake information to the public.
USGS Pasadena Field Office and the
provide earthquake information for
Southern California. As is usually the case, these events caused large
increases in the traffic to our web servers. The typical pattern shows a
large spike which peaks about 10 minutes after the event,
an exponential decay in the traffic rate.
The first event occurred at 18:26 PST. It was apparently felt over a wide
populated area. The traffic on our web server began increasing
almost immediately. This can be seen in
Figure 1. These graphs show the overall traffic on the USGS and
Trinet web servers. The peak recorded after the first event was
90 hits/sec on the USGS server, and 25 hits/sec on Trinet.
This peak was starting to decay when the second event
occurred at 18:50 PST. The second event's traffic peaked at 134 hits/sec
on the USGS server, and 53 hits/sec on Trinet. Shortly after,
the Trinet server experienced another surge after the URL for
the Shake Map was shown on TV. This peak reached 233 hits/sec,
and is the largest traffic surge we have recorded as of this writing.
The brown line in the graph is the ratio of cache hits to
total activity on the Squid server,
which is the reverse-proxy server in front of the main web server.
The main pages that people come to see after an earthquake are the
Internet Intensity Maps pages. Recent Earthquakes shows a
near-real-time map of California with all events for the past week plotted.
The CIIM pages show maps of perceived shaking felt by the public, and
include an online questionnaire for people to fill out to report their
experiences. A detailed look at the overall traffic for two hours around
the events is show in
Figure 2. The double peak can be seen clearly here.
Figure 3 shows traffic
for just the Recent Earthquakes pages. The overall pattern looks about
the same. The main thing to note is that this traffic amounted to nearly
half of the total activity on the USGS site.
Figure 4 shows the calls
made to ciim_update.pl. This is the CGI script that is called
after a visitor has filled out the online questionnaire. It writes a summary
of their input for later processing and provides them with an estimated
Mercalli Intensity for their location. There are two features of note in this
graph. First is that the peak activity was on the order of 84 questionnaires
per minute. This shows that the new server is performing well. Before
we got the new server, there were
problems with processing questionnaires after the September 3
Yountville earthquake. The new server did not have any apparent problems
dealing with the load after this event, even though the load was nearly
the same. In particular, the 'submitted' and 'processed' lines on the
graph match exactly, except for one questionnaire submitted at 19:02.
This request failed with a
which was most likely due to faulty input. The second feature to note in
this graph is that the peak
for the second event was delayed. For some reason, the CIIM map
and questionnaire for the second event was not created until about
30 minutes after the event. This is the reason the second peak is
Another popular page is the
Shake Map. This is served off of the
Trinet web server.
shows activity for the Shake Map pages after the events. The
top graph shows all HTTP activity for the /shake/ directory
tree, and the bottom graph shows just page views for the
two events on Saturday. The
interesting feature here is the third peak on the first graph.
The first two peaks
correspond to the traffic peaks after each of the two events.
The third peak came about two minutes after KCBS news
showed the shake map and URL on TV. After this, the Trinet
web server recorded 233 hits/second for several minutes. This
was the largest traffic spike to date on our web servers.
The data rate on the Trinet Squid server was about 6Mb/sec
at the peak, and the server CPU was about 50% idle. This
corresponds well with our experiences in testing, which indicated
that a single K6-2 Squid server should be able to handle on the
order of 400 hits/sec.
|Total shake map activity
|Shake Map page views
Overall, it appears that the current web server configuration performed
well under load. The lessons from the Yountville earthquake helped us
to successfully keep up with the load this time. Still, the continued
growth of the Internet means we will probably have to further increase
our capacity in the future.
17 January, 2001
- Stan Schwarz
- Honeywell Technical Services
- Southern California Seismic Network Contract
- Pasadena, California