Internet Reliability and Outages
the 4:11 Effect: The Power Failure and the Internet, IEEE
Security & Privacy, 2003.
frame relay SLA
frequencies between SLAC, FNAL CMU and CERN
seconds betwen SLAC, FNAL CMU and CERN
on Best-effort IP Networks, Les Cottrell, SLAC (ITU-T SG13/SG16
Workshop on IP Networking and Mediacom 2004, Geneva, Switzerland, April
- Risk of
internet collapse rising (Nov. 26, 2002)
competency -- ISP backbones stand up in grueling 30-day performance
- Two providers - C&W and Savvis - delivered picture-perfect
availability. Both had zero downtime, with Savvis running trouble-free
for the full monthlong test. C&W also had perfect uptime, but its test
window began a few days later than other providers because of a test
configuration error on its part.
- The networks of four providers - C&W, Level 3, Savvis and WilTel - met
or exceeded the vaunted “five nines” standard for network uptime
during normal operations.
- Sprint’s numbers for average delay were the theoretical minimum rates
for a beam of light traveling cross-country (see story).
- Average jitter for all ISPs was measured in microseconds, well below
the point where application performance could suffer.
- Packet loss for all providers averaged just 0.01%.
Six Five-Nines?, BCR, April 23, 2002.
Network Is Vulnerable, Report Finds, August 6, 2002.
Infrastructure Vulnerability Assessment: Background and
Recommendations (NANOG June 2002)
Unimaginable Emergency Put Communications to the Test, NYT, Sept.
That day, AT&T handled 431 million voice calls - 20 percent more than
usual and the most it had ever carried on a business day.
Computer Snag Renders Credit Cards Useless for a Day
Outages - Internet Performance Measurement and Analysis (IPMA)
- Abilene network
outages, with description of cause and duration
gives its guarantee to network availability: Under the terms of the
company's availability guarantee, MIS customers are guaranteed
99.5-percent network availability. If a customer's service is down for
15 minutes or more per day, AT&T will credit the customer for one to 10
days of service. MIS' monthly service fee is priced from $595 to
$2,095, depending on access speed.
power: flywheel backup power
Phone System Reliability
- AT&T network
At AT&T's Network and Computing Services organization, one of the
most important gauges of network reliability is Defects Per Million.
This measurement is a statistically valid record of how many calls per
million did not go through the first time because of a network
procedural, hardware or software failure. Defects-Per-Million is not an
average, it is an accurate accounting of network performance that is
tallied by the day, as well as by month-to-date and year-to-date.
During 1997, AT&T's Defects-Per-Million performance was 173, which
means that of every one million calls placed on the AT&T network, only
173 did not go through the first time due to a network failure. That
equals a network reliability rate of 99.98 percent for 1997.
- FCC requires reports of outages affecting 30,000 or more customers
for more than 30 minutes. "According to Bellcore, 972 outages have been
reported since ... 1992." (America's Network, Nov. 15, 1997)
- "[A]bout 50% of outages impacting more than 30,000 customers for
more than 30 minutes have been caused by facility outages, and more than
half of these outages are caused by fiber cable dig-ups. About 30,000
customers a day have lost access to the public switched network for an
average of five hours because of [facility outages]." (America's
Network, July 1, 1998)
outages in the telephone network leave about 30,000 customers a day
for an average of 5 hours without telephone service, where about 50% of
facility outages are the result of fiber cable dig-ups. Detailed outage information
- "Approximately 30 to 40% of all outages are attributable to human
intervention. The other root causes include software faults, hardware
failures and a number of miscellaneous reasons, including
weather-related problems. Outages are also caused by the rush repair
jobs necessary to remedy these situations."
Internet Technical Notes and Resources
by Henning Schulzrinne