If you’ve followed the M-Lab blog over the past year and a half, you’ll know that we’ve been engineering and testing a massive upgrade to our server software and how it’s managed and deployed    . That’s been a ton of work, and our team is excited to now complete the migration from the PLC/vserver platform to the new k8s stack.
We expect some great derivative outcomes of migrating to Docker and Kubernetes both for ourselves in managing the platform and measurement tests we host, and for the community as well. In particular, the NDT server was completely rewritten in Golang, and being fully “dockerized” the server is now portable and usable outside of the M-Lab platform by others. This is really exciting because not only can M-Lab use our tools to measure the public Internet, but anyone can run our server now to measure their network privately using the same methods and tools. This post kicks off a series of blogs that will demonstrate various ways to use M-Lab server and client tools, starting with
We deployed the new M-Lab platform to 1/3rd of the M-Lab fleet, and now we need to assess whether or not it is a performance regression, relative to the old platform. As long as we can be sure the performance of the new platform does not constitute a regression, then we can roll out the new platform and be confident that we have not made anything worse.
The first annual Michigan Broadband Summit was held on Sept. 24, 2019, sponsored by the MERIT research and education network, and it was my pleasure to attend on behalf of M-Lab. The gathering was an opportunity to learn more about the amazing work that MERIT, industry, municipalities, cooperatives, school districts and others are doing to improve internet access and service quality in Michigan. I came away inspired by every conversation and presentation, with the keynote from Francella Ochillo from Next Century Cities, and the MERIT team’s presentation of the Moonshot framework being particularly moving.
While we’ve been hard at work this year on the M-Lab 2.0 Platform Upgrade and Global Pilot, the number of people all over the world runing NDT tests has continued to grow. We collected 2 billion total NDT results between 2009-01-01 and the beginning of the second quarter of 2019, and we are on track to add 500 million just from April through September of this year! We now regularly exceed 3 million NDT tests per day, compared with 2.4 million per day at the end of the first quarter of 2019, 1 million per day two years ago, and 50k per day four years ago.
To make our traceroute data in BigQuery more useful, researchers have sought an easy way to reconstruct the path of hops for the same test. This task was particularly hard because the schema, which was designed many years ago, put the hops of the same test in different rows.
To address this need from many of our partners and researchers, M-Lab is delighted to announce that the
traceroute BigQuery table in the
aggregate dataset is now available to the public. The new
traceroute schema has one test per row, and all hops for a single test are inside the same row.
- Stock Linux 4.19 LTS kernels with modern TCP and Cubic congestion control
- Standard instrumentation for all experiments using tcp-info
- Virtualization and container management using Kubernetes and Docker
- Reimplementation of the NDT server
M-Lab is working on replacing the current traceroute BigQuery table with new schema, which will put all hops of one test in one row of BigQuery table. The new table will have all the information in the current table but make the search of hops within one test much easier. To make this happen, we will stop the new data feed of current traceroute BigQuery table in early July, 2019. The details of new schema will be published once the conversion of all data to BigQuery tables with the new traceroute schema is completed and available to the public.
Earlier this year, M-Lab published blog post outlining our new ETL pipeline and transition to new BigQuery tables. That post also outlined where we’ve saved our datasets, tables, and views in BigQuery historically, and recommended tables and views for most researchers to use. At that time we also implemented semantic versioning to new dataset and table releases at that time, and began publishing BigQuery views that unify our NDT data across multiple schema iterations and migrations.
M-Lab is excited to highlight the launch of “Michigan Moonshot”, a pilot data collection project developed through a partnership with the MERIT Network, Michigan State University’s Quello Center, and M-Lab. We worked with MERIT developers who adapted the Piecewise codebase to build the Michigan Moonshot website.
If you’ve clicked on the Visualizations page on our website in the last year or so, and tried to find recent data for any location, you’ve likely been disappointed.
TLDR; - Our visualization site is BACK! Continue reading to learn more..
Last year, we outlined our plans to Modernize the M-Lab Platform. This year, we’re bringing them to life. Here’s a summary of why the platform update is so valuable and what you can expect throughout the year.
AFRICOMM 2018. Left to right: ￼Amreesh Phokeer (AFRINIC)￼, Josiah Chavula (University of Capetown), Georgia Bullen (M-Lab), Antoine Delvaux (perfSonar), Stephen Soltesz (M-Lab).
In late November 2018, M-Lab was invited to the Internet Measurement Workshop at AFRINIC-29 in Tunisia and to give a keynote about M-Lab and open internet measurement at AFRICOMM 2018 in Senegal. Both trips were a fantastic opportunity to deepen our relationship with researchers focused on the African Internet, learn more about how our platform is serving community needs, foster conversation around open Internet measurement, and identify opportunities for further collaboration, research and tool development to better support the Internet measurement, research and policy community in Africa.
M-Lab had the pleasure of attending the first ever SIGCOMM hackathon on August 25, 2018, at the Nokia Skypark headquarters in Budapest, Hungary. The hackathon, sponsored by Nokia, DECIX, and Netflix, invited network research faculty, students, and industry professionals from around the world to form teams and develop tools, new features or analyses during the Saturday following the SIGCOMM conference.
We’ve reached a point in human history where, for many of us, the Internet has become a standard presence in our daily lives. In the United States, the Internet is simply part of how many of us engage with the world. In other countries (and parts of this one), the Internet remains unaffordable, unreliable, and inaccessible. The Internet unites us in many ways, and at the center of work on the future of the Internet is a dedicated community of experts exploring the questions that will move the Internet to the next level of its evolution: What is an open Internet? What is a healthy Internet? What factors contribute to the Internet ecosystem’s health?
On February 1st, 2018, during a regular data quality review, we identified an increase in switch discards at sites with 10Gbps equipment connected to 1Gbps uplinks. We used our switch telemetry data to assess whether there were any negative consequences for tests contained in our SideStream or NDT data sets, and then we used the same data sets to determine whether our remediation strategy had any negative effects. In both cases, we found no observable effects, indicating that everything was below the noise floor for Internet performance data.
- When: Saturday, August 25, 2018
- Where: SIGCOMM, Budapest, Hungary
- When: Aug. 7, 2018 - Aug. 8, 2018, 9AM - 5PM
- Where: New America, 740 15th St NW #900, Washington, D.C. 20005
Measurement Lab is turning 10! On August 7 and 8, we look forward to gathering the Measurement Lab community to showcase how the platform has evolved, learn from you about how you are using M-Lab, and discuss how we plan for the next 10 years of measuring the Internet and providing public data to the world. So much has changed over the last 10 years (and that’s not just our expanding volume of longitudinal data!), come celebrate, brainstorm, analyze, and share with us.
Since June 2016, M-Lab has collected high resolution switch telemetry for each M-Lab server and site uplink.
Originally designed to detect switch discards from server traffic microbursts, we now support the DIScard COllection (a.k.a. DISCO) dataset as a standard M-Lab BigQuery table:
The Measurement Lab team has always tried to make it as easy as possible to run network measurements. Currently, most users run tests either directly from the M-Lab website, or through a 3rd party integration. Over the years, many users have requested the ability to run tests on a regular basis, e.g. daily or weekly to collect data over time. Today, we’re releasing a tool that will help you do just that.
As a platform committed to producing empirical data for the public, Measurement Lab (M-Lab) has historically supplied regulators and other governmental entities with technical facts pertinent rule-making processes. In our February 2015 submission to the FCC’s Open Internet docket, we committed to research on the state of broadband and performance impact of interconnection in the United States. Earlier this year, the FCC began the process of re-evaluating its authority over broadband Internet services, and opened a Notice of Proposed Rulemaking. This blogpost is a shortened version comments that M-Lab filed in the docket regarding its continued research on the impact of interconnection on consumer broadband. The full filing in the FCC docket includes an elaboration of our research with additional supporting evidence and charts.
M-Lab data is collected from distributed experiments hosted on servers all over the world, processed in a pipeline, and published for free in both raw and parsed (structured) formats. The back end processing component for this has served us well for many years, but it’s been showing its age recently. As M-Lab collects an increasing amount of data thanks to new partnerships, we have been concerned that it will not be as reliable.
In February 2017, M-Lab was notified of issues with the M-Lab data available in BigQuery. Upon investigation, a problem was identified with the Paris Traceroute collection daemon which resulted in a reduction in Paris Traceroute measurements beginning in June 2016. At the peak of the outage, fourth quarter 2016 - January 2017, approximately 5% of NDT tests had an associated Paris Traceroute test. Additionally, an issue within the data processing pipeline resulted in Paris Traceroute data that was measured and collected, not being inserted into the BigQuery tables and therefore available for use.
The Measurement Lab (M-Lab) global measurement platform contains the largest collection of open Internet performance data on the planet. Hundreds of thousands of people (and rising everyday!) test their Internet connections on M-Lab and provide that data to the public domain for research, analysis and advocacy. As a consortium of research, industry, and public-interest partners, M-Lab is dedicated to providing an ecosystem for the open, verifiable measurement of global network performance. With an extensive library of data dating back to 2009, M-Lab has over 5 petabytes of information about quality of experience on the Internet contributed by people like you.
The M-Lab team has been working with Bocoup’s Data Visualization team to overhaul our visualizations and give all of you better support in exploring all of the Measurement Lab data. Look for more about that soon – and reach out via email@example.com if you are interested in helping with testing and user feedback!
M-Lab has lots of data, but depending on how you slice the data, you might end up with too small of a sample size on a given day in a given location. As part of designing the Observatory visualization, we came up with the idea of using dotted lines to show that we didn’t have a large enough sample size to assert the data value, but that leaving out the data would be incorrect as well. The Bocoup Team took that further and developed a new d3 plugin, d3-line-chunked, which allows you to easily visualize gaps in your data and has good animation support.
In August 2015, M-Lab was notified of potential degradation of site performance by a measurement partner based on discrepancies compared to results for their own servers. After a full investigation these patterns were found to have been caused by the unique confluence of several specific conditions. Interim remediation measures were taken in early October 2015, and the resolution of the degradation was confirmed by the partner and others. Due to these administrative actions, the episode, which we are calling the “switch discard issue,” has not affected testing conducted in the United States (the region impacted by this problem) since October 11, 2015, and thus measurements after this period are not affected by the incident. M-Lab has also conducted an evaluation of data collected during the time period in which the issue occurred, and has taken steps to remove affected measurements from its dataset. This incident will not affect use of its dataset, past or present, as a result.
In January, M-Lab launched a beta test of new BigQuery tables for M-Lab data. Today, M-Lab is pleased to announce that the beta test was successful. The new, faster-performing tables will be M-Lab’s new standard BigQuery tables.
Before we move on to specifics, when we say faster performing, we mean a lot faster. As in, certain queries that used to take over 2 hours now complete in 8 seconds. That means that playing with the data just became a lot more fun.
To help users dig in to this data as quickly and seamlessly as possible, M-Lab has consolidated all of its data documentation and updated it to show how to take advantage of the new tables.
Today, M-Lab is happy to announce the public beta of new M-Lab BigQuery tables. These tables provide substantially improved performance and reduce the difficulty of writing BigQuery SQL.
The team working on archiving M-Lab data recently discovered that the M-Lab data hosted in BigQuery was affected by a bug that caused duplicates to appear in our dataset. Queries against M-Lab’s BigQuery dataset performed between May 2014 and April 2015 were impacted. The raw files in our Google Cloud Storage bucket were not.