![[ANT logo]](../../images/logos/ant_logo_1_128px.png){kind=logo}
Internet Outage Datasets
This web page documents our datasets about Internet outages. Our datasets are available upon request.
What data do you have? We have 24x7 data since Oct. 2014, as well as supplemental data (ASes, geolocation, etc.) to support analysis of outage data.
What is in the data? We document the format of outage datasets. (If you’re studying how we geneate this data, we also have “intermediate” outage datasets.)
Where to get started?
We recommend using outagedownup data for most purposes,
since it includes post-processing cleans up some known flaws
that occur in the raw data.
Outagesdownup format data is in the “alL” datasets, so
for example internet_outage_adaptive_a58all-20241001 has outagdownup data
for 2024q4.
You may want to use outages format if you want to see when different observers see different things. Outages data is also in the “all” datasets.
If you want ot see the individual probes each observer sends,
then you need outageraw format.
Outageraw data is provided separately for each site, so
one would request internet_outage_adaptive_a58w-20241001
and internet_outage_adaptive_a58c-20241001
to get two of the outageraw datasets that went in to a58all.
Most Recent Datasets
We release updated data every quarter. Our most recent datasets are listed at our datasets page. Researchers are encouraged to reqeust datasets.
We show near-real results in our outage website.
In addition to quarterly datasets, we have an API to access near real-time data. Please contact John Heidemann for access to streaming data.
Datasets on Our Comparisons of Batch and Streaming Outage Detection
We have compared batch and streaming versions of Trinocular. Streaming Trinocular provides near-real-time updates in our outage website, but the demands to report results quickly mean that some algorithms cannot be sued there, such as algorithms that require a multi-day window of data. Batch Trinocular includes many algorithms, including those we have added to handle corner cases. We have compared these batch and streaming version of Trinocular, finding that the two similar systems agree most of the time, more than 84%. We show that only 0.2% of the time the algorithms disagree, and 15% of the time only one reports. [1]
- Erica Stutz, John Heidemann and Yuri Pradkin 2025. Quantifying Differences Between Batch and Streaming Detection of Internet Outages. Proceedings of the IFIP International Workshop on Traffic Monitoring and Analysis (Copenhagen, Denmark, Jun. 2025), to appear. [PDF] Details
This work used Trinocular batch processe data from 2021q1 and 1 week of streaming data starting 2021-02-26. We expect to make the streaming data available by paper publication in 2025-06.
| name | shortname |
|---|---|
internet_outage_adaptive_a43all-20210101 |
2021q1 a43 |
We are currently preparing the streaming data for release and expect it to be available before paper publication in June 2025.
Datasets on Our Work Understanding ISP Address Dynamics
We have explored ISP address dynamics (how they move customers around in their address space), and used that information to improve accuracy of outage detection. [1]
- Guillermo Baltra, Xiao Song and John Heidemann 2024. Ebb and Flow: Implications of ISP Address Dynamics. Proceedings of the Passive and Active Measurement Workshop (Virtual Location, Mar. 2024). [PDF] [Dataset] Details
This work used data from outage data from USC/ISI and data from RIPE Altas.
The ANT datasets are:
| name | shortname |
|---|---|
internet_outage_adaptive_a30all-20171006 |
2017q4 a30 |
internet_outage_adaptive_a42all-20201001 |
2020q4 a42all |
internet_outage_adaptive_a42w-20201001 |
2020q4 a42w |
internet_outage_adaptive_a42n-20201001 |
2020q4 a42n |
The RIPE Atlas measurement IDs were: 1010.
This work generated some new datasets:
| name | shortname |
|---|---|
internet_outage_maintenance_a42all-20201001 |
2020q4-maintenance |
ip_accumulation_as_a42-20201001 |
2024q4 per-AS accumulation |
ip_accumulation_prefix_a42-20201001 |
2024q4 per-routable-prefix accumulation |
Datasets on Our Work Improving Outage Detection
We have developed two algorithms for improving accuracy in outage detection. We gather more information for sparse blocks to reduce false outage reports, and we detect maintenance activity using only external information. [1]
- Guillermo Baltra and John Heidemann 2019. Improving the Optics of Active Outage Detection (extended). Technical Report ISI-TR-733. USC/Information Sciences Institute. [PDF] Details
Our work uses dataset A27 to analyze Iraqi government mandated outages:
| name | shortname | Start Date | Duration | /24 Blocks |
|---|---|---|---|---|
internet_outage_adaptive_a27w-20170101 |
2017q1 A27w | 2017-01-01 | 92 days | 4,070,885 |
internet_outage_adaptive_a27g-20170101 |
2017q1 A27g | 2017-01-01 | 92 days | 4,070,885 |
internet_outage_adaptive_a27j-20170101 |
2017q1 A27j | 2017-01-01 | 92 days | 4,070,885 |
internet_outage_adaptive_a27c-20170101 |
2017q1 A27c | 2017-01-01 | 92 days | 4,070,885 |
| Iraqi blocks of 2017q1 | 2017q1-Iraq | 2017-01-01 | 92 days | 1,176 |
Our work uses dataset A30 to analyze CenturyLink (AS209) address renumbering:
| name | shortname | Start Date | Duration | /24 Blocks |
|---|---|---|---|---|
internet_outage_adaptive_a30w-20171006 |
2017q4 A30w | 2017-10-06 | 87 days | 4,033,972 |
internet_outage_adaptive_a30g-20171006 |
2017q4 A30g | 2017-10-06 | 87 days | 4,033,972 |
internet_outage_adaptive_a30e-20171006 |
2017q4 A30e | 2017-10-06 | 87 days | 4,033,972 |
internet_outage_adaptive_a30j-20171006 |
2017q4 A30j | 2017-10-06 | 87 days | 4,033,972 |
internet_outage_adaptive_a30c-20171006 |
2017q4 A30c | 2017-10-06 | 87 days | 4,033,972 |
internet_outage_adaptive_a30n-20171006 |
2017q4 A30n | 2017-10-06 | 87 days | 4,033,972 |
| CenturyLink (AS209) blocks of 2017q4 | 2017q4-AS209 | 2017-10-06 | 87 days | 35,935 |
Datasets on Our Work Clustering Outages
We have developed two algorithms for clustering outages, anycast catchments, and routing information. We cluster by linear ordering for visualization, and the other event clustering to group blocks by events over time. These algorithms are described in our technical report [1]
- John Heidemann, Yuri Pradkin and Aqib Nisar 2018. Back Out: End-to-end Inference of Common Points-of-Failure in the Internet (extended). Technical Report ISI-TR-724. USC/Information Sciences Institute. [PDF] Details
Our clustering work uses the following datasets:
| name | shortname | Start Date | Duration | /24 Blocks |
|---|---|---|---|---|
internet_outage_adaptive_a17all-20140701 |
2014q3 A17all | 2014-07-01 | 92 days | 4,034,614 |
| the 172/8 subset of 2014q3 | 2014q3-172/8 | 2014-07-01 | 92 days | 6,415 |
| RIPE Atlas J-Root CHAOS | J-Root | 2015-11-30 | 1 day | 9305 VPs |
Datasets to Studies of Hurricanes Harvey, Irma, and Maria (2017)
We have studied Hurricanes Harvey, Irma, and Maria (2017) in the Hurricane Harvey web page.
This analysis uses data from a29
internet_outage_adaptive_a29all-20170702.
Datasets on Diurnal Networks
For the following papers [1] [2]
- Lin Quan, John Heidemann and Yuri Pradkin 2014. When the Internet Sleeps: Correlating Diurnal Networks With External Factors. Proceedings of the ACM Internet Measurement Conference (Vancouver, BC, Canada, Nov. 2014), 87–100. [DOI] [PDF] Details
- Lin Quan, John Heidemann and Yuri Pradkin 2014. When the Internet Sleeps: Correlating Diurnal Networks With External Factors (extended). Technical Report ISI-TR-2014-691b. USC/Information Sciences Institute. [PDF] Details
and use the following datasets (A12w, although we also evaluated A12c and A12j):
| name | shortname | Start Date | Duration | /24 Blocks |
|---|---|---|---|---|
internet_outage_adaptive_a12w-20130424 |
A12w | 2013-04-24 | 35 days | 3703717 |
internet_outage_adaptive_a12c-20130424 |
A12c | 2013-04-24 | 35 days | 3703717 |
internet_outage_adaptive_a12j-20130424 |
A12j | 2013-04-24 | 35 days | 3703717 |
Datasets from Our Paper on Trinocular
For this paper [1] we use the A7 dataset, containing 3 sites, ISI (w), CSU (c) and Japan (j):
- Lin Quan, John Heidemann and Yuri Pradkin 2013. Trinocular: Understanding Internet Reliability Through Adaptive Probing. Proceedings of the ACM SIGCOMM Conference (Hong Kong, China, Aug. 2013), 255–266. [DOI] [PDF] Details
| name | shortname | Start Date | Duration | /24 Blocks |
|---|---|---|---|---|
internet_outage_adaptive_a7w-20130212 |
A7w | 2013-02-12 | 2 days | 3648487 |
internet_outage_adaptive_a7c-20130212 |
A7c | 2013-02-12 | 2 days | 3648487 |
internet_outage_adaptive_a7j-20130212 |
A7j | 2013-02-12 | 2 days | 3648487 |
We have also collected a similar, but much longer month long dataset, A12:
| name | shortname | Start Date | Duration | /24 Blocks |
|---|---|---|---|---|
internet_outage_adaptive_a12w-20130424 |
A12w | 2013-04-24 | 35 days | 3703717 |
internet_outage_adaptive_a12c-20130424 |
A12c | 2013-04-24 | 35 days | 3703717 |
internet_outage_adaptive_a12j-20130424 |
A12j | 2013-04-24 | 35 days | 3703717 |
As described in the paper, our system is currently running 24x7. Subsequent doutage datasets are listed on our general list as datasets with “outage” in their name.
Dataset A12 was also used in analysis of the diurnal internet.
Datasets to Studies of Hurricane Sandy (2012)
We have studied Hurricane Sandy first in a 2012 technical report [1] then later in the Trinocular paper [1] and presented this work in several talks [3] [4] [5].
- Lin Quan, John Heidemann and Yuri Pradkin 2013. Trinocular: Understanding Internet Reliability Through Adaptive Probing. Proceedings of the ACM SIGCOMM Conference (Hong Kong, China, Aug. 2013), 255–266. [DOI] [PDF] Details
- John Heidemann 2013. Long-term Data Collection and Analysis of Outages at the Edge. Talk given at CAIDA Workshop on Active Internet Measurement Systems. [PDF] Details
- John Heidemann 2013. Active Probing of Edge Networks: Outages During Hurricane Sandy. Talk given at NANOG57 as part of panel hosted by James Cowie. [PDF] Details
- John Heidemann 2013. Active Probing of Edge Networks: Hurricane Sandy and Beyond. Talk given at FCC Workshop on Network Resiliency. [PDF] Details
- John Heidemann, Lin Quan and Yuri Pradkin 2012. A Preliminary Analysis of Network Outages During Hurricane Sandy. Technical Report ISI-TR-2008-685b. USC/Information Sciences Institute. [PDF] Details
This analysis uses the raw data in Internet Survey 50j.
Dataset input and results are at:
input: USC-LANDER/internet_address_survey_reprobing_it50j-20121027.output: USC-LANDER/internet_outage_survey_it50j-20121026.
Datasets from our Technical Reports On Outage Detection
Earlier work on outage detection (pre-trinocular) was in this technical report: and revised analysis is in the Trincular paper The technical report includes some unique datasets.
- Lin Quan, John Heidemann and Yuri Pradkin 2013. Trinocular: Understanding Internet Reliability Through Adaptive Probing. Proceedings of the ACM SIGCOMM Conference (Hong Kong, China, Aug. 2013), 255–266. [DOI] [PDF] Details
- Lin Quan, John Heidemann and Yuri Pradkin 2012. Detecting Internet Outages with Precise Active Probing (extended). Technical Report ISI-TR-2012-678b. USC/Information Sciences Institute. [PDF] Details
This work analyzes:
- Regular surveys: S29w (2009-11-02) to S44j (2011-12-05), each a 2 week continuous probe on 22k /24 blocks. See technical report for details.
| Survey | Start Date | Duration | /24 Blocks | Analyzable |
|---|---|---|---|---|
| S29w | 2009-11-02 | 14 | 22371 | 10389 |
| S29c | 2009-11-17 | 14 | 22371 | 10085 |
| S30w | 2009-12-13 | 14 | 22381 | 10629 |
| S30c | 2010-01-06 | 14 | 22381 | 10853 |
| S31w | 2010-02-08 | 14 | 22376 | 10788 |
| S31c | 2010-02-26 | 14 | 22376 | 10876 |
| S32w | 2010-03-29 | 14 | 22377 | 10750 |
| S32c | 2010-04-13 | 14 | 22377 | 10807 |
| S33w | 2010-05-14 | 14 | 22377 | 10701 |
| S33c | 2010-06-01 | 14 | 22377 | 10727 |
| S34w | 2010-07-07 | 14 | 22376 | 10623 |
| S34c | 2010-07-28 | 14 | 22376 | 10610 |
| S35w | 2010-08-18 | 14 | 22376 | 10591 |
| S35c | 2010-09-02 | 14 | 22375 | 10585 |
| S36w | 2010-10-05 | 14 | 22375 | 10679 |
| S36c | 2010-10-19 | 14 | 22375 | 10733 |
| S37w | 2010-11-24 | 14 | 22374 | 10633 |
| S37c | 2010-12-09 | 14 | 22373 | 10647 |
| S38w | 2011-01-02 | 14 | 22375 | 10598 |
| S38c | 2011-01-27 | 14 | 22373 | 10553 |
| S39w | 2011-02-20 | 16 | 22375 | 11585 |
| S39c | 2011-03-08 | 14 | 22375 | 10955 |
| S39w2 | 2011-03-22 | 14 | 22374 | 10904 |
| S40w | 2011-04-06 | 14 | 22922 | 10794 |
| S40c | 2011-04-20 | 14 | 22921 | 10874 |
| S41w | 2011-05-20 | 14 | 40645 | 23065 |
| S41c | 2011-06-06 | 14 | 40639 | 23092 |
| S42w | 2011-07-26 | 14 | 40565 | 21259 |
| S42c | 2011-08-09 | 14 | 40566 | 22723 |
| S43w | 2011-09-13 | 14 | 40598 | 21361 |
| S43c | 2011-09-27 | 14 | 40597 | 22784 |
| S43j | 2011-10-12 | 14 | 40594 | 22055 |
| S44w | 2011-11-02 | 14 | 40634 | 22971 |
| S44c | 2011-11-16 | 14 | 40632 | 23084 |
| S44j | 2011-12-05 | 14 | 40631 | 22581 |
- W1: 24 hour probe (2011-09-28). internet outages it1w-20110928 format
- W~2: 24 hour probe (2012-08-03). Internally named as
internet_address_super_survey_reprobing_it49w-20120803. Details to come. - W~3: 24 hour probe (2012-08-09) from 3 sites. Internally, they are
named as
internet_address_super_survey_reprobing_it49w-20120809,internet_address_super_survey_reprobing_it49c-20120809,internet_address_super_survey_reprobing_it49j-20120810. Details to come.
Getting this data
For the detailed formats of each dataset, please refer to the corresponding README file at our datasets.