30 Comments

Summary:

For most people, 4G feels a little faster, but not anything close to the 10 times we were promised. What’s going on here? Ed Robinson of Riverbed Technology thinks we’ve made websites so obese that the networks can’t keep up.

just a shutter trick

We’ve all heard the commercials and seen the ads. The major cellular carriers are rolling out their shiny new 4G networks, and they’re promising speeds “up to 10 times faster” than regular 3G networks. They also toss the term “LTE” around as if it were going out of style.

Cellular marketing hype has only added to the confusion. For instance, Apple’s iOS 5.1 update changed a descriptive label in the iPhone 4S UI from 3G to 4G, and nobody is quite sure if it means anything.

For most people, 4G feels a little faster, but not anything close to the 10 times we were promised. With all the misinformation emanating from the industry, it’s about time that we brought a dose of reality to the conversation.

A few years ago, I was leading a startup that was building a collaboration application in New Zealand. The load times for the app were so slow that the app was unusable, and the issue threatened the viability of the company itself. After digging into the problem we found that networks were performing as well as they could, and we had to find another way to speed up the app.

Ultimately, we built a solution that optimized the content itself for better performance, rather than focusing on network and hardware acceleration. The result was so effective that we started a new company, Aptimize, just to sell the acceleration software. We realized at the time that everyone was focusing on networks, but the real problem was the content traveling over those networks.

And today, that’s precisely the area where we can make the largest performance gains — content optimization. All this content — from Twitter share buttons to Facebook Like buttons all the way down to images and videos, new analytics packages, interactive ads and third-party comment systems — has made websites so obese that the networks can’t keep up.

Back to reality, oh there goes gravity

If 4G were legitimately 10 times faster than 3G, I’d never have to buy another phone to keep up with technological advances again.

Think I’m kidding?

Let’s say a site loads in 7 seconds on your iPhone over 3G. That means the site should load in less than a second over 4G every time, right?

Wrong.

In real-world use cases, the promise of 4G is almost never a reality. The network itself is absolutely capable of performing 10 times faster than 3G networks. That part is true.

But Web performance isn’t that simple. There is a huge delta between pure science in a lab and the real world.

The cellular network, or your ISP’s local network, is a fraction of the infrastructure and services that it takes to get sites to load in your browser. In fact, the network that your device connects to is the last mile in the Web performance continuum.

It’s not the second-to-last mile. It’s dead last.

While we’re all excited about the arrival of 4G speeds, consumers need to realize that 4G can also deliver speeds up to 10 times slower than 3G, or dial-up, depending on real-world variables.

One major difference between 4G and 3G is that a 4G network is all IP-based, so it passes information between phone and carrier exactly the same way as a Web browser communicates with the Internet. 4G also boasts an increase in theoretical maximum throughput from 14.4 Mbps (mega bits per second) to 100 Mbps or more.

But theoretical maximums are like the theoretical miles-per-gallon on a new car. They’re nothing like what you experience day to day. In the real world, a 4G network delivers maybe three to five Mbps, and 3G networks two to three Mbps. This is due to distance from the cell tower, interference from walls and objects, and people moving while using their device — real world things.

Carriers have been fine-tuning their 3G networks to account for these real-world variables for years. Because 4G is newer, it hasn’t had anywhere near this much tuning yet.

In addition, the “throughput” is not your connection to everything on the Internet, as might be assumed. It is actually your connection to the carrier. These are two different things. Most of us aren’t going to buy a 4G phone so we can access AT&T’s data center faster. Instead, we want increased speed while we do real things on the Internet.

Try a test: the average Web page is about 500 KB in size. If a 4G network is operating at 100 Mbps, the Web page should download to your phone in four-hundredths of a second. That’s about the time it takes to blink (500 KB is four megabits and in theory the network is operating at 100 megabits per second).

You don’t need a stopwatch to know that this is almost never the case.

Keep your eyes on the ball

Perhaps the most frustrating thing about 4G marketing is that it causes otherwise intelligent people to ignore logic and reason.

Even worse, we’ve been through this before. Back when content delivery networks (CDN) were the hot new technology, we put all of our Web performance eggs in that one basket. We behaved as if they were going to save the Web and usher in a new era of light speed connections. Similarly, as server hardware grew in both processing power and affordability, we assumed that Web performance was finally solved.

All of these technologies are important, and any time we make an advance in one area, we need to remind ourselves we’re not done yet. We need to collectively refocus our attention on a holistic view of performance, so we can make progress that outpaces any single breakthrough. This is the only solution.

Below is a breakdown of the various components that exist between the server hosting a website and your device that affect how fast a site might load:

The Web performance continuum

Whenever you access a website, or download an app, the request gets routed through an array of computers and services all working together. Loading a Web page doesn’t just access the array once, there are literally 50 or more requests that go backwards and forwards. Any speed problems get amplified when you’re dealing with this quantity.

One of the biggest determinants of how fast this works is not the client, gateway or Web services, but the connections between them — how fast the pieces can talk to each other. This is often known as latency — high latency means a slow connection. Latency is higher for wireless connections (WiFi or cellphones) and grows with distance. The further away you are physically from the website, the longer it takes to load.

Your upgrade to a 4G phone, or high-speed Internet only affects the connection between you and the gateway. If the bottleneck is somewhere else in the system (it often is), then the website will load at the speed of the slowest part of the network. It’s like driving to work. No matter how well paved your driveway is, a traffic jam will still slow you down.

Don’t just tap the piñata 

When you hit a piñata, you don’t just tap it. You hack at it as hard as you can, and you aim for the spots that were already weakened by other people’s blows.

My point?

When it comes to Web performance, you need to invest in the areas that have the highest likelihood for significant returns. The networks are already fast enough. We need to find other areas that promise more return on performance investments.

By using techniques such as compression, HTTP request merging and dynamic layout effects to optimize content, we can make vast improvements in real-world speeds and load times. The opportunity in this area of acceleration can return as much as 40 to 50 percent improvements in load times.

Sure, that’s not 10 times, but it works in the real world. And unlike the claims the 4G guys are plastering on billboards, it’s a promise we can believe in.

Ed Robinson is the general manager of Web content optimization at Riverbed Technology. He was the co-founder of Aptimize, a web acceleration company that was acquired by Riverbed in 2011.

Some rights reserved by vinaykr.

  1. Jeff Kibuule Sunday, April 8, 2012

    Downloading a file and rendering a webpage aren’t the same thing if you are trying to compare the speed of a connection.

    Share
    1. Nigga, then will you please define what’s new definition of connection speed?

      Share
      1. What’s your problem, boy?

        Share
    2. Q3 technologies Sunday, April 8, 2012

      Sorry to break this to you Jeff, but they are pretty much the same thing.

      Share
      1. No they aren’t, file download requires one DNS request (if any at all), webpages require at least one DNS request and quite posssibly many more. Webpages can require decompression, third party code and extensive rendering in the browser. A file download has none of these.

        Unequivocally they are not anywhere near the same thing.

        Share
      2. No they are not the same thing. A web page requires some device power(therefore wait time) to display the downloaded page. Jeff has a good point here.

        Share
  2. The download speeds absolutely are 10c better. Upload even more so. The question is what you are using your connection for. In my experience the bottlenecks come from rendering issues

    Share
  3. When are software programmers going to stop blaming the networks and hardware when their app doesn’t perform as fast as it should and learn to optimize their code. Being an Ops guy it always amazes me when I see the sloppy code that they’re putting out.

    Share
    1. Absolutely right, there are so many craplications written by clueless developers, it makes us network guys point and laugh.

      Share
  4. Very informative article, thank you very much!

    Share
  5. Ed Robinson Sunday, April 8, 2012

    pcvip, you’re right – the rendering issues are a common bottleneck. Most consumers don’t understand this, nor should they have to – they see 10x speeds and think its the only solution they need. Ed

    Share
  6. On my 4G LTE LG Revolution on Verizon, living in the SF Bay Area, and using the Speedtest.net app… I get 28mbps download and 24mbps upload. I’d say that’s 10 times faster.

    Share
  7. Having worked in telecom in the old POTS days, I would guess that cell towers send requests to a collection point. Could the overall process be speeded up by having a DNS database replicated at each of these points so DNS lookup could be done locally?

    Share
    1. DNS is done locally in lte

      Share
      1. Can you please define ‘locally’. As in / on the user’s device / in my bedroom / at the cell tower / at the downtown switching center / at the border of the network?

        Share
  8. My Droid Bionic gets 20-30 mbps all the time on verizon. that might not be 10 x but that’s definitely more than what you have posted

    Share
  9. Warren Whitlock Sunday, April 8, 2012

    speed ratings should not have to be felt. There are objective tests that measure the claims carriers make so that’s easy to debunk.

    The other things mentioned here are real.. but when it comes to downloading a file, watching a video, etc. download speed is all that counts.

    Share
  10. It is ironic that one of the sites I visit most, is also one of the worst designed from a data perspective. Dear Gigaom, please clean this place up!

    Share

Comments have been disabled for this post