Measuring and Understanding Broadband: Speed, Quality and ...

7 downloads 3219 Views 158KB Size Report
Jun 8, 2010 ... That Ookla looks beyond the concept of the “speed test” and towards .... Small binary files are downloaded from the web server to the client to ...
Measuring and Understanding Broadband: Speed, Quality and Application June 8th, 2010

There are a variety of efforts underway that should help the FCC understand the state of broadband, and the FCC/Sam Knows program is merely one of the latest. Although unsolicited, Ookla recently released Net Index, a free compilation of results broken down by geographic regions, derived from the more than 1.5 billion test results performed using our technology. Those performed at our free sites, Speedtest.net and Pingtest.net, are directly utilized, and combined, these sites provide a free and accurate way to quickly measure every important technical aspect of a broadband connection – download and upload speeds (throughput), packet loss, jitter and latency (ping) – that are widely accepted as the fundamental attributes necessary to a quality Internet experience. That Ookla looks beyond the concept of the “speed test” and towards true service quality makes us a unique and important voice in the ongoing education of residential broadband customers, academic thought leaders, and industry stakeholders. Before diving into the methodologies of our testing and why measurement matters, let’s examine Ookla’s background, and address concerns that our free services and data are anything but serious technical tools and resources from a profitable, thriving, and sustainable business. Our paid licensing division counts almost every major Internet Service Provider (ISP) in the world as a client, including more than 80 percent of the top ISPs in the U.S. plus hundreds internationally. A wide variety of interested parties all around the globe use our solutions, including the FCC. Our applications and data already serve a wide range of paying clients in the academic (Harvard, MIT, Stanford), service (AT&T, Comcast, Cox, Verizon, Vonage), hardware (Nokia, Siemens, Sun, Linksys), and media (Reuters, Doubleclick, CNN, ESPN) markets. So how did we become the single largest and most trusted source for broadband measurement? As founder and CEO, I built Ookla, on a foundation that includes 10 years of personal leadership at Speakeasy, a broadband service provider sold to Best Buy in 2007, and the collective team experience of 30 years working at a national independent broadband provider. We pioneered a focus on faster upload speeds, an emphasis on lower latency connections to enable voice-over-IP (VoIP) and were recognized for our tireless effort to deliver a richer broadband experience for our customers. This background provides team Ookla with a unique and deep set of experiences with which to build the applications that allow millions of daily visitors to determine if their providers are measuring up. We deeply understand that, in the presence of competition, there is tremendous incentive to deliver more, for less. With this unique perspective and compounded experience, it’s safe to say that we know better than most that to understand the quality of a connection, you must go beyond simply measuring speed. Broadband measurement is a highly variable process. For example, testing to and from a server far away introduces latency that, in turn, means a slower download speed. Testing to a completely

different server demonstrates results that vary based on “network distance,” all while reflecting the real-world fact that the Internet itself, beyond an Internet Service Provider’s (ISP) network, is unpredictable and mostly unmanageable. For this reason, our Net Index utilizes only those test results where the server is within 300 miles of the end-user, largely eliminating the significant latency introduced when transiting multiple “hops” (i.e. routers) and minimizing the “Internet middle” significantly. Still, people want to test their service quality not just across town, but also to servers around the world and quite understandably for this is ultimately how a connection is used. We have shifted from a hyper-local to a daily, global Internet experience, accessing information from servers scattered all over the world. We understand this basic fact, allowing people to select any of 600 servers at Speedtest.net understanding that, when they do, those particular results are naturally going to vary widely. The real-world network context must be considered when evaluating the accuracy of any test being applied to it. So with this in mind, it’s obvious you must first understand what you’re measuring before interpreting the results. It’s complicated, measuring the various aspects of a connection isn’t a perfect science nor is it as simple as many people would like. Frustration and confusion are the unfortunate side effects of complex problems and variable measurements, and this is no exception. However, when this confusion bleeds into and muddies the conversation about the National Broadband Plan, the risk to the evolution of broadband in the United States is great. What follows is an honest and open exploration of the methodology, data, and analysis that will provide a framework for next steps of the broadband revolution happening in America today.

Guiding Principles Of Ookla Broadband Measurement 1.

What We Measure

2.

How We Measure What We Measure

3.

Why We Measure What We Measure

What We Measure

Reference: http://blog.ookla.com/2010/05/14/testing-speed-tests/

We measure the maximum sustainable throughput of a connection at a given location to and from a server of their choosing. We believe we do this better than anyone else for two reasons: 1.

In some cases, other speed tests are not trying to measure this but rather, as an example, the speed you might see when streaming a video clip or downloading a PDF via a web browser. Measuring a single file transfer is entirely different than measuring maximum sustainable capacity of the connection.

2.

Our testing engine is superior for this task due to specific technical reasons we discuss in our blog.

Remember that, with many variables in play, even two tests run back to back to the same server can and do vary somewhat. Whether it’s an application updating itself in the background or your DVR downloading an HD trailer for a VoD (Video on Demand) service, occasionally a test result is “wrong,” and it’s for this reason that we encourage people to test their connection multiple times and at different times of the day. The results are always correct, but only insofar as they reflect the potential of the connection at that moment in time. With Net Index, and in all of the broadband reports we produce, we critically analyze results from the same computer (using a combination of IP address and cookie to identify each uniquely) in order to mitigate these types of issues.

How We Measure What We Measure

Reference (again): http://blog.ookla.com/2010/05/14/testing-speed-tests/

For the sake of avoiding significant technical details that can make this process unnecessarily difficult to comprehend, here is a brief outline of our download testing methodology, once initiated by a client: 1.

Small binary files are downloaded from the web server to the client to estimate the connection speed

2.

Based on this result, one of several file sizes is selected to use for the real download test

3.

Up to 8 parallel HTTP threads are employed for the test, based on the estimated speed

4.

Throughput samples are received at up to 30 times per second

5.

These samples are then aggregated into 20 slices (each being 5 percent of the samples)

6.

The fastest 10 percent and slowest 30 percent of the slices are then discarded

7.

The remaining slices are averaged together to determine the final result

Cache prevention and other technical fail-safes, as well as environmental corrections and remedies are employed throughout the testing process. Eliminating portions of the test results are done to eliminate ramp-up and other anomalies, this method is the result of intensive testing and provides optimum accuracy. The test is a true measurement of HTTP throughput from the server to and from your client. The server controls packet size and all other TCP parameters, so tuning is performed to maximize throughput potential. No method or analysis is left unexplored when it comes to pursuit of the most accurate result. One glaring problem with many “state of the Internet” reports is that businesses, schools and other nonresidential locations are included in their results, which creates rankings that are highly misleading when attempting to determine, as is usually the case, the state of household broadband. Currently at Net Index, Ookla includes and ranks only those tests that, after multiple sophisticated filters, are determined to be from a residence. For many years some reports have cited Sandy City, Utah and Iowa City, Iowa as two of the fastest cities in America. These are the homes of the University of Utah and the University of Iowa – two campuses that have, as is true at most major universities now, implemented not only extremely fast Internet connections that aren’t realistic choices for most people at home, but also web caching servers that deliver a lot of content over an Ethernet or Fiber LAN, the speed of which appears to be included in these historical reports. What you measure is key, and Ookla is careful to ensure it’s strictly that measurement that can and should be used to compare to the “headline” or “promised speeds” of residential users that are the subject of current discussion and debate in the U.S.

Why We Measure What We Measure Reference: http://blog.ookla.com/2010/05/10/why-take-a-speed-test/

With a firm understanding of what should be measured and why, read more about why you should join the 40 million people who use our speed and quality tests every month. It’s also a method that allows you to understand the true potential of your broadband connection rather than only showing you what a single application on a single computer can do. This reflects the practical reality of the connected home where if it’s not due to multiple applications it’s multiple computers and increasingly, a wide variety and number of devices that require Internet connectivity to function. In the popular

plumbing analogy, we measure the equivalent of the maximum flow of water feeding your home rather than the flow of a single faucet. In terms of measuring and managing broadband for the purposes of understanding what an ISP is providing you, this is the only method that accurately reflects the critical components of download and upload speed. While alternative testing methods do exist and loosely use the term “speed test” to describe their application, those results are naturally going to vary from ours. This is a part of what creates confusion in the marketplace, leading some analysts and technical professionals to conclude “all speed tests vary widely” and, to paraphrase, “produce results that are all over the place and can’t be counted on.” Ookla firmly believes that, once you understand what we’re measuring, you will also understand why we’re measuring it and can be confident that the results you see are absolutely accurate1 and actionable. With a better understanding of what should be measured and why, read more about why you should join the 40 million people who use our speed and quality tests every month, 40 percent of them new users.

1  In rare circumstances some security software can act as a proxy, delivering the speed test payload from cache, meaning from the local memory on the client machine. In such cases, users see a result that is obviously highly unrealistic. We eliminate such “bogus” records from our reports and over the years have eliminated almost all of these problems. We don’t believe that asking someone to turn off their anti-virus software is a reasonable solution and fortunately, the next iteration of our application makes web proxies like these an entirely moot issue.

A Closer Look 1.

Inconsistencies

2.

Misconceptions

3.

Key Considerations

Inconsistencies As discussed earlier, variance in the questions being asked - not to mention the methods employed attempting to answer them - lead to significant inconsistencies. These inconsistencies in turn lead to misunderstanding and completely false conclusions by trusted authorities on the subject of broadband performance. Most often, these false or damaging statements are completely accidental. In other cases, they are convenient straw man arguments necessary to support foregone conclusions. In still others, they impact the very nature of the discussion, throw people off the trail of what is so critical to pursue and too often, lead to “subject fatigue” where people simply begin to lose interest due to on-going confusion and heated, dead-end arguments. It is the responsibility of stakeholders to consider areas where they can uniquely add value, even if merely to corroborate results. Any contribution has merit and should be aggressively explored for the value it contains when the very future of a country’s broadband infrastructure is at stake. The Internet represents the single greatest technical achievement in the history of mankind. At the same time, in relative terms, we are just beginning to discover what a ubiquitous reliable, high-speed communications and digital delivery system can do for people, business, institutions, and states. The stakes are very high, critically impactful to our nation’s future GDP, to say nothing of the countless benefits society as a whole. Ookla believes that the FCC should be using every resource at its disposal to achieve the aggressive and worthwhile goals that have been set out. To miss opportunities to make use of a massive data store that promises untold depth of information regarding the state of the Internet in the United States is unacceptable, much more so when the alternative carries a multi-million dollar cost and a questionable track-record when previously deployed in other countries.

Misconceptions Some may question the validity of the data made freely available by Ookla, going so far as to label it unscientific (due primarily to self-selection) and therefore useless to policymakers. This vocal minority goes on to suggest that asking some 10,000 individuals, just 200 per state, to allow a third-party and foreign owned company to deploy mysterious devices on their home network, is somehow much more so (scientific). Let us explore that claim. We feel compelled to point out that asking for volunteers willing to add a device to their home network that is going to monitor everything they do on the Internet will, without a doubt, create a

common demographic and psychographic profile, not a broad-based representation of the Internet. Here we see self-selection in play in a more pronounced and impactful way than at Speedtest.net, and with a sample set that is but one hundredth of one percent (0.0158 percent) of the 65M unique U.S. visits to our Speedtest and Pingtest sites in just the past 12 months. Of great concern, participants will be required to meet certain criteria, which will likely eliminate extremely key participants in a study such as this. Consider for a moment that one of the requirements to be a volunteer and accept this device into your home is that you must not be a heavy user (download more than 30 GB of traffic per month2). First-hand experience dictates that it is precisely these heavy users that compound the challenges of any broadband ISP’s network management, but we would encourage anyone doubting the claim to speak with the operators at other major carriers. One could argue that finding out exactly what it is these people are up to should be one of the primary goals in determining whether the infrastructure, packages, and policies (for starters) are geared toward a workable future. Finally, we understand the study is to go on for 25 months and, so, if you are likely to move or change providers in that time, you are also undesirable. This all but disqualifies residential users in the younger, more mobile, demographic of Gen-Y, who are more prone to changes of residence and service due to economic and personal upheaval. It is these users who realize the heaviest and most forward-thinking access of broadband service in the U.S. and will be the key audience served by any improvements stemming from the FCC’s current efforts3. The concept of “science” is being liberally applied here, on face value, those claiming this method vastly superior to the data freely available around both quality and speed, plus the data the FCC is collecting using the MLabs and Ookla testing engines are either well outside of their range of expertise, or severely misinformed.

Key Considerations There are a number of key elements involved in creating and operating a global broadband testing solution, what follows are a few that set us apart:

1.

Volume of Data • Ookla compiles more than a million speed and line quality test results every

2  A single Blu-ray movie is between 8-12 GB and a standard DVD at least 4 GB. Apparently if you watch more than one or two movies per week over the Internet you aren’t a material part of the future of broadband in the United States.! The streaming and VoD industry, not to mention the future of television is just took a big hit if so. 3  Gen-Y moves residence and employer every 12-18 months, on average. http://www.census.gov/population/ www/socdemo/migrate/cal-mig-exp.html & http://www.businessweek.com/investor/content/jun2007/ pi20070624_294649.htm

day, dating back to September of 2007, history of 5 percent compound growth each month since. • Data collected includes the entire range of technologies, wireline (cable, DSL, fiber) and wireless (WiMax, Microwave) and major mobile devices (iPhone, Android) across the entire globe. 2. Testing Server Infrastructure • Proposed alternatives for speed and quality testing is limited to just eight servers and in some cases “encouraging ISPs to also host nodes within their network,” resulting in measurements of strictly the “last-mile” and unable to account for Internet backbone capacity and congestion. • Severely limited test server infrastructure results in significant latency due to distance that will dramatically impact the results. • The Ookla server infrastructure has been in place for many years and continues to add new servers almost daily. More than 600 servers are in service, including a presence in every country recognized by the U.N. Ookla offers more than 140 servers in the United States alone. 3.

Fully Managed, Dedicated Host Infrastructure • Ookla uses rigorous QA measurements to verify servers are performing accurately based on daily server statistics, reports from users and geographic comparisons to other servers in the area.

4.

Compatible, Immediate and Free • Our web-based applications work across all Operating Systems and all major Browsers. • There are no special requirements for user data collection, such as mandatory hardware or user profiles. • Pingtest.net and Speedtest.net are free tools that anyone can use.

5.

Globally Accepted and Comparable Solution • Because Ookla collects and compiles data from around the world using the same technology employed by thousands of ISPs, it is consistent, making it accurately comparable to a huge number of other data sources.

6. Ookla Is An Industry-Adopted Standard • Serving thousands of ISPs, ASPs and MSOs, we are in a unique position to understand and measure their networks, our tools are constantly evolving in accordance with changes in the broadband industry. • The FCC and several other U.S. Government Agencies, along with Independent Non-Profits, and other BB Initiatives use our solution. • Approximately 257 million unique individuals have used Speedtest.net or Pingtest.net to test their broadband performance. 7.

User Privacy • All tests are completed voluntarily by users and any private information is removed from Public Data and Reports. • No installation, software or hardware, is required, so there is no opportunity for a third party device to monitor traffic or activity on the user’s computer or network.

Why You Should Care To be clear, it’s going to take more than one study or data resource to get a clear understanding of residential broadband service in the United States. It is easy to talk about the limitations or issues involved in one method or another without bothering to recognize that, with a little effort and thought, many of these can be largely mitigated or effectively eliminated and extremely valuable data derived. It is in this spirit that we explore some of the common sense issues that ought to be a part of the conversation. A recent survey (PDF) that reported 80 percent of people don’t know what their speed is supposed to be could actually be quite misleading. For one thing, it was a question asked via a telephone survey where you are expecting the person to know the answer off the top of their head. It is easy to imagine most people would need a minute to look at their bill or check their online statement. Many people you ask don’t know exactly how much they are paying each month for cable. How many people know the total amps of power being delivered to their home, the number of BTUs their furnace creates, or the horsepower of their car? For most people, most of the time, their broadband service works perfectly well for what they use it for and even if they experience problems or slowdowns, the answer doesn’t lie in memorizing their broadband speed – most don’t really care and shouldn’t have to. Since when do we expect consumers, on a broad basis, to know the answers to the technical attributes of even fundamental home services and products they use every day? It’s as if the Internet is so new we believe there’s a

unique way of measuring and managing it. Why are we making this so difficult? Consider this series of uncontroversial, actionable questions that exemplify the sorts of things we really need to know as we look to our broadband future: A survey recently carried out at Speedtest.net resulted in more than 100,000 unique completed entries in less than one week. Importantly, this approach eliminates the variable of self-selection in terms of its impact on the data – it does not matter whether a person is technically inclined or not when we already know their connection speed and the result of speed tests performed on that connection. Indeed, even the geekiest among us are limited to choose from the packages provided by the ISPs in our area.

What do you do with your broadband connection in the mornings? email/web upload media

stream video download media

work

Does that differ between what you do with it in the evenings or on weekends (your spare time)? email/web

stream video

upload media

download media

work

What are the biggest problems you have with your connection:

usually slow only certain times of the day complete disconnects, "freezes", all services stop working

Not Important

Somewhat Important

Important

Very Important

What do you pay for your broadband? Do you get everything you want for that price, if not, what is it lacking? Yes

No

Would you pay more for that, if so, how much? No

Yes

Rank (1-5) the aspects of your broadband that are the most important to you: Price

Speed

Line Quality

Services

Value

Please rank from1-7, your complaints about your broadband service currently: (1 being the biggest complaint) Slow Periods

Price

Support

Download Speed

And so it’s important not to miss the Upload Speed Services Line Quality point of these exercises. People want to know if they are getting what they pay for. Governments want to know if carriers are sufficiently competitive such that a healthy market is in play and just about everyone, carriers included, are eager to successfully deploy next-generation technologies and perform the infrastructure upgrades necessary to keep customers happy. Consider all of the variables in play, measuring the connections of just 10,000 Americans, whether you do that occasionally or all day long for a year, seems like a flimsy way to exclusively shape broadband policy. So what is a more valuable way to measure, and ultimately address the questions the FCC is struggling with in the ongoing debate? For starters, Ookla has made over 800 million speed and quality test records available to the FCC, for free. As of today, Ookla now politely asks Speedtest.net visitors to enter the speed they see on their bill and how much they pay for this “promised speed” from their service provider. This information will allow us to compare specific package offerings from an ISP in a given area and report on not only what people are paying per Mb of speed but also detail how closely providers come

to meeting or beating expectations. This is merely the beginning of a continuing series of indexes backed by a massive data set in collaboration with major academic institutions and our clients, all aimed at delivering a faster and higher quality Internet for the community we serve. Our goal is not to shape government policy4, but to frame the debate surrounding these questions with the help of facts and a free exchange of information. After all, isn’t that the very spirit and promise of the Internet, itself?

@ copyright 2010 Ookla

4  Full disclosure: We do not want nor are we applying for any future government work in this area but some of the data we have collected historically and several of our applications are already deployed and serving the FCC and other federal government agencies today.