Comcast and Net Speed Tests

JimDaGeek writes "I recently moved to Columbia, SC where I have Time Warner as my cable ISP and pay for an 8 Mbps connection and have been very happy with the service, speed, and reliability. In contrast I have heard bad things about Comcast. So now that I am up in the Philadelphia PA area visiting my parents, I decided to test out the speed and reliability using the Speakeasy speed test. The results surprised me. Here are the reported download speeds in Kbps: New York, 18,946; Washington, 15,821; Atlanta, 11,257; Chicago, 10,042; San Francisco, 4,230. What is going on? I know my father is not paying for a 10+ Mbps connection. Is Comcast giving priority to popular speed-test sites?" From Comcast's site, in the Philadelphia area they seem to offer download speeds of 6 or 8 Mbps, with an option for a "PowerBoost" to 12 Mbps on large files. This wouldn't explain the results JimDaGeek got of almost 19 Mbps down.

Update: 07/10 12:07 GMT by KD : A friend in Massachusetts had a tree fall on his house. The Comcast guy who reconnected the lines told him that they are boosting the line speed to 20 Mbps down / 2 Mbps up in certain areas to be more competitive with Verizon FiOS.

Comments

SpeedTest.net

[ragnarok]

The Speakeasy speed test is just a re-branded version of speedtest.net. They have a lot more test locations to choose from there.

Re:SpeedTest.net

[Lumpy]

and it's a worthless test.

One big transfer != bandwidth capabilities.

Give me 100-500 smaller files with smaller ACK going back. that gives you a real test that will show latency and jitter.

also check many different ports. Port 80 get's priority. Ports above 8000 get lower priority. Ports for Voip are screwed with hard.

Re:SpeedTest.net

[Anonymous Coward]

bandwidth != latency. You want both (high bandwidth and low latency, that is), but that doesn't make them equal.

Time of day?

[carlivar]

Since cable bandwidth is shared, wouldn't the time of the test matter? I've noted (very unscientifically) that my Internet seems slower between roughly 7-9pm (on Charter in Los Angeles area).

Re:Time of day?

[catwh0re]

It also makes me think of how complicated the simple "bandwidth test" will be should net neutrality get thrown out the window.

Already we juggle the factors of location, "paid for speed", shared bandwidth issues with daytime or peak traffic.. but then without some kind of neutrality we'd also be juggling whether or not the interconnects between yourself and the test site are all on a higher priority or lower priority pipe.(something we could never know)

Today your ISP can blame a bit of the slowdowns on network conditions, but ultimately it's obvious if your ISP is a slower provider.. but in the future they'll be able to knowingly serve you slow speeds while claiming it's just the low-priority sites you may be visiting.

Re:Time of day?

[kaufmanmoore]

Facebook and myspace can eat bandwith with all the crap some people post.

Re:Time of day?

[lancejjj]

Since cable bandwidth is shared, wouldn't the time of the test matter? FYI, practically all internet traffic traverses shared communication lines. The point-to-point connection of DSL becomes shared as soon as your local copper pair is squeezed onto a multiplexed line, within a few thousand feet of your home.

Re:Time of day?

[MrShaggy]

Since cable bandwidth is shared, wouldn't the time of the test matter? I've noted (very unscientifically) that my Internet seems slower between roughly 7-9pm (on Charter in Los Angeles area). I look at the difference between cable and dsl as 2 different ways of getting to a train-station.


Cable might be like taking a bus to the station. There might be other people there. However, its not that significant.


Dsl is like taking your car or a cab to the station. It definitely is not a shared ride into the station.


The one thing that happens is that everyone gets off of there rides, and they all take the train out. Pretty much at the same speed.

Of course during that time of the day there are many people on the internet, so it won't matter what you are using. Its somewhat insignificant how 'slow' it is. The only thing that kills the dsl is how far away you are from the office. Thats pretty wild. The thing is that you can get speed tests from eith side of the country or even planet. So even if it isnt the rush hour here, it will be elsewhere.


Thats why I think most dsl ads are so misleading.

Re:Time of day?

[Firehed]

You remind me of the time that I single-handedly saturated half the pipe for the entire school with a couple of unusually fast torrents. Oh, what I wouldn't do for a personal T3...

Re:Time of day?

[Matey-O]

Oh, what I wouldn't do for a personal T3...

What, short of paying money for it?

Re:Time of day?

[ivan256]

Yes, and your cable connection uses a shared upstream connection once it reaches the CO too. The difference is that the "last-mile" connection is also shared. Either can become a bottleneck depending on traffic.

Of course the internet traverses shared lines. That's practically the point of the internet.

Re:Time of day?

[Short Circuit]

That's nothing. I was in a vacant computer lab in Michigan during Spring Break, pulling Debian CD images from an FTP server in Germany (best transfer rate I could get; Jigdo never worked for me, and there weren't official torrent trackers at the time. {Are there now? Switched to Ubuntu a couple years back.})

The door to the lab bursts open, and a couple techs started methodically checking port labels on all 69 of the PCs, then started checking the Macs. (I was on a Mac 'cause those were the only machines that had burners.) They let me finish downloading disc 12, but told me not to start disc 13.

In 2003, it was kinda noticeable when someone had three simultaneous 200KB/s FTP transfers running. Fortunately, I wasn't in violation of the Acceptable Use Agreement, as I hadn't installed any unauthorized software...

Re:Time of day?

[Short Circuit]

this is because even if programs aren't using bandwidth networks use a csma/cd "carrier sense multiple access with collision detection" setup where, simply put, if someone is transmitting on the line, you back off and wait to transmit. While that's technically in the Ethernet standard, you don't see it much--if at all--with switched networks, and even less with networks with decent switches that hold packets in a queue. Switches and routers aim to eliminate that kind of inefficiency.

Re:Time of day?

[CopaceticOpus]

What, short of paying money for it?

Dude, a T3 would be almost as expensive as just buying all those CDs and movies.

:)

Giving Comcast Props

[tenchiken]

It kills me to say something nice about the brood of bloodsuckers that are Comcast, but I can verify that Pockets of Comcast's net are seeing huge increases. In particular, I have seen speeds of 19-22Mb/s burst to testing sites, and almost 2.0MB/s non-bursting.

That's in the Denver region using both speedtest.net and DSL tools.

Give credit where it's due, but Comcast does appear to be amping up the bandwidth hugely.

Between this and the Zimbra announcement, Comcast has firmly passed Qwest as next to last evil corporation.

Re:Giving Comcast Props

[Eric in SF]

Two things:

I have Comcast (8/768) in Cole Valley, San Francisco, and I have also noticed the speed increase. Uploads to my website are now cruising at 140kb/sec, occasionally dropping to 90kb/s. No complaints here! I performed a dslreports speed test recently and it also reported some Korean or Scandanavian-class bandwith numbers - the highest I've certainly ever seen in my time with broadband.

Second - it's my understanding that as you saturate the uplink connection (max out uploading a file) on a consumer-grade connection/router, you interrupt the normal control-channel "Chatter" of web browsing. Basically, the "I got it" packets are stuck due to the saturated uplink, and you don't get the next packet until the acknowledgement makes it.

I could be completely wrong - I am by no means a networking expert, so if this is wrong, be gentle.

Re:Giving Comcast Props

[Blkdeath]

Second - it's my understanding that as you saturate the uplink connection (max out uploading a file) on a consumer-grade connection/router, you interrupt the normal control-channel "Chatter" of web browsing. Basically, the "I got it" packets are stuck due to the saturated uplink, and you don't get the next packet until the acknowledgement makes it.

Actually you're absolutely right. Any TCP (connection based) protocol will suffer from a saturated uplink to the point where it can become unusable depending on how greedy that client is about your outbound packet flow for exactly that reason; every time you request a packet you have to ACKnowledge its receipt. If enough packets go unacknowledged the sender will stop transmitting to give you a chance to catch up. At the point where you have your outbound 100% utilized, your connection becomes useless for even the simplest of tasks.

The reason residential connections tend to be asymmetrical is the fact that ack packets are much smaller than received data and the nature of the Internet nowadays (even with our media rich content) is that people are downloading files in the 100s of megabytes (a 720p movie trailer on Apple's website will generally run about 120 meg) but only need to send 1/10th or less back to the sender to acknowledge receipt.

Some time ago my room mate got heavily into filesharing networks. Upon trying out a new client he found he was getting better speeds. Well, that's because his download speed was based on a ratio to his upload speed and the limits I'd set in his old client naturally were not set in the new one.

I tried to connect to my home server from work to check my e-mail and found that the connection was slower than a 14.4k dialup modem. I ssh'd in, found the source of the traffic and promptly shut down all direct-to-Internet traffic from his machine and poof! No more link saturation!

Re:Giving Comcast Props

[Firehed]

As a slashdotter, I doubt it, but could you have confused Mbps with MBps?

When I was on Comcast (actually, come to think of it, I think I am here too), I usually got speeds right around what was advertised. Hell, I got a speed boost when they took over Adelphia.

Re:Giving Comcast Props

[dreddnott]

I've been using dial-up at home for all of ten years. They won't even pave our frontage road let alone string cable or DSL out here.

How do I do it? Well, I suppose I'm just used to it. I'd say I use Opera to cope, as it's such a snappy, efficient browser, but I've been using Opera for at least eight years (version 3.something), since before the Internet was bloated, so I guess that doesn't count. I do set the Opera caches to their maximum sizes to minimize the horror of redownloading the static content of websites I visit frequently. Ah yes, I remember now. I really started using Opera because the download size was (and still is) extremely small compared to the other browsers.

I never go to Youtube, don't use MySpace, Google Maps, or other silly websites that break if you don't download the content fast enough. I browse Slashdot with more minimalist settings, with my threshold set at -1 and -all- posts displayed inline so I don't have to click on any comments to actually read them.

On other web forums I usually use the simplest scheme (Relicnews forums has a great setting called "Adminimal") and if I have trouble on a particular site or I get impatient (usually after half an hour of waiting) I'll turn image downloading off and manually download the images that I think I'll need to see.

Google displays 100 search results per page, and I set all forums to the maximum number of posts per page. Display Print Version is a welcome feature on those retarded hardware review websites. It's much better if I download it all in one go, there's a lot of overhead for me in CSS/HTML these days.

On the whole it's not so bad, over the last decade I've still managed to acquire a substantial collection of MP3s (I was in fact one of the 300,000 names on Metallica's hitlist for Napster). Dial-up first started to frustrate me in 2001 when I started making anime music videos. I used my own DVDs as source material but uploading the videos was quite a pain. I remember the first time I downloaded a large file on dial-up, I think it was around 32 megabytes. I was very proud of myself.

I used to play Unreal Tournament over the Internet when the game was very popular (1999-2002). I think the best ping I ever got was around 150-160, and I was so used to playing at over 250ms that I won every round on that server. I was very competitive at LAN parties (ok, I won grand prize at a tournament for budding game developers) but rarely got the opportunity to play at one. I think I could reclaim the glorious days of my youth if I had a better Internet connection...or if this podunk wasteland they call the High Desert of Southern California actually got some good local events going on.

Well thank god we still have TW in Raleigh

[gelfling]

What with their rock solid 3.7MB down and 374KB up - Anything more would be Goddamn communism.

DOCSIS 2.0 Plus

[Fallen Kell]

It all depends on the cable modem that you have. Some of the new motorola modems, like the SB5120, do not have the ability for Comcast to limit as much as they sometimes like. Comcast themselves has not been too worried about it as long as the network segment you are a member of is not over-crowded. They see it more like a new benefit which allows better competition against FIOS. Personally I average 25-28mbps on my modem.

Here is a current snapshot:

PowerBoost uses compression

[rdean400]

I'm wondering if the speed test uses a data block that is more compressible.

PowerBoost

[the_cowgod]

The Speakeasy test regularly reports over 20Mbps on my Comcast connection. The "PowerBoost" feature allows basically uncapped speeds for the first ~15MB of a transfer, then it drops down to the normal 6Mbps. I can easily see this effect when doing large downloads with my UsenetServer account. It does inflate speed test results, but Comcast does not appear to be favoring the test sites in any way.

Download a linux distro

[Raleel]

Someone above mentioned using jigdo to get all of debian, but even using wget to get a full dvd or cd set of some distro will provide you with a good data point. That's been my standard test for quite a few years now.

Re:Download a linux distro

[grcumb]

Someone above mentioned using jigdo to get all of debian, but even using wget to get a full dvd or cd set of some distro will provide you with a good data point.

I beg to differ. Downloading a single file is only indicative of how fast a particular connection is between two particular points. P2P, if it's allowed to, saturates the network with two-way traffic to numerous end points. If - and this is a big if - there were no constraints on P2P traffic at large on our networks, I would consider it a near ideal measure of TCP traffic capacity in the real world.

In fairness to your comment, you characterised that single download as a 'data point'. Strictly speaking, that's a valid statement. How to get enough of these data points to provide useful insight isn't adequately answered by either of our suggestions, and frankly I'm a little doubtful about the online speed tests being discussed here, too.

Do It Yourself

[Doc Ruby]

The Speakeasy speed tests are indeed easy, and easy to "speak" about on their site with posted ratings. But there's nothing magic about it, that you couldn't do with simple commands from your PC.

All you've got to do is fire up a shell (whether Windows, Linux, or other client OS), and download a big (>10MB) file while timing it. Find an HTML link to a video or something, then download it from the shell (eg. wget or curl in Linux) to a local directory. Watch the minutes and seconds from when you first connect (right after you give the command, after you get the download feedback), to when the file is complete. Then examine (eg. ls on Linux, or use your GUI file manager) the file for its exact size in bytes, then divide the time by the size.

I know this seems obvious, but distrusting Speakeasy's numbers as cooked by Comcast shouldn't be the last act before punting to Slashdot. Real tests, not just examples like Speakeasy, are trivial to run by yourself.

Lots of factors...

[SuperBanana]

To name a few:

• Router/signal compatibility. Comcast has switched signals several times and in two cases, it caused horrendous performance. There's also a particular cablemodem (has a big blue "sleep" button on top) that has serious problems with high packet rates and connections (ie, BitTorrent crashes it.)
• Interference.
• Local loop, backbone, and uplink utilization. Guess what, guys? Sometimes traffic peaks for strange reasons. Sometimes it's a virus outbreak, a new movie trailer, or a big news story.
• THE INTERNET . It's unreliable, not guaranteed, never has been, and YOU ALL KNOW THIS AND HAVE BEEN TOLD IT OVER AND OVER AND OVER. I work for a university. We have seriously fat pipes. We have a 10Gbit backbone. And some sites I can FTP from at 2MB/sec. Others, I get 40KB/sec. "Speed tester" services compete for bandwidth just like everyone else. Stop holding them up as some pure, holy source of internet traffic that magically flows through every router at maximum speed.
• Some content is akamai-zed. I get Apple's software updates at maximum line speed, for example.

If you're not happy with your service, CALL THEM. My parents were some of the first people to get MediaOne service back around '98-'99, and every time they had problems, we picked up the phone, and it was taken care of.

I've had the same experience elsewhere. Any time I have problem with the service, be it regular disconnects or lousy performance- I pick up the phone, and a few minutes later someone is checking into signal to noise ratios and such. If you lease the modem, they're usually happy to try sending out a tech and swapping out a modem if you're polite but clear there's a problem. They're usually even more amenable if you pick up the modem yourself at a "service center."

In my years as a customer and having friends who were customers, I've seen a)flooded junction boxes b)in-house distribution amps turned up too high c)1 failed modem d)one buggy model e)several incompatible modems after "upgrades" to the area network (usually to support faster speeds.)

In short: call comcast, ask them to look into it. They've almost always been helpful, through all the various company changes: MediaOne, RoadRunner, etc.

Re:Lots of factors...

[pokerdad]

In short: call comcast, ask them to look into it. They've almost always been helpful, through all the various company changes: MediaOne, RoadRunner, etc.

As someone who used to work for Comcast allow me to say rotflmao. Either you are one lucky sob or you are lying.

Just as a matter of example (one among many) during the entire nine months I worked for Comcast the entire state of Illinois never left the outage board. That isn't to say that no one in Illinois ever had a connection, but many people had little or no connection and we were under instruction to do absolutely nothing for anyone from Illinois - just keep BSing them till they gave up.

20 Meg Cable Modems Practical Since 1999.

[Dave Burstein]

19 megabits on an unloaded cable modem node is quite practical, especially for brief speed tests.The U.S. cable modem architecture is based around nodes that share 35 megabits or so downstream. RCN even sells a 20 megabit service, which David Reed buys and reports gives him the promised 20 megabits most of the time. In practice, most nodes run with 5-10 megabits typical load, as only a fraction of the time multiple users on one node simultaneously download megabits and fill the pipe. Statistical multiplexing (sharing) works much better than people expect on cable modem networks, especially on well managed nets that split nodes and otherwise expand capacity when needed because a node gets frequently congested.

Comcast and cable suppliers are working on several techniques to allow customers to get more than the 6 or 8 meg typically allocated, while not causing undue congestion. "Speed burst" technology tests the network load, and if uncongested temporarily raises the speed of an individual modem making a fast download for a brief period. While that's marketed as "doubling" speeds to 12 and 16 megabits, bursts to 20 and 25 megabits are also practical.

The new technologies require upgraded equipment and are typically being tested first and then rolled out market by market. So it would be no surprise if a subscriber in Philadelphia (Comcast's home town) is benefiting from a test or early deployment of faster speeds than Comcast customers elsewhere.

100 megabit+ (shared) cable modems are being deployed in Japan, Quebec, and France, bonding 3 or 4 35 megabit channels for higher speeds. These are early "DOCSIS 3.0" products, unlikely to be widely deployed in the U.S. until 2009. Comcast's CEO, Brian Roberts, demonstrated 100 megabits at the cable show in Las Vegas this spring, and will probably test widely in 2008 and go into deployment (especially where Verizon is building FIOS) the following year. DOCSIS 3.0 requires a new cable modem unit, however, so this customer is unlikely to be an early tester.

That doesn't explain why the test to San Francisco only ran at 4 megabits, which could be explained by node congestion a few minutes later, inferior Comcast backbone connections to Speakeasy's host in San Francisco, or other circumstances. For more details on coming faster cable modems, google DOCSIS 3.0.

Dave Burstein

Editor, DSL Prime

They are up to it.

[WindBourne]

I tested my comcast 6 MB with and got 1.2 MB., but With speakeasy, it ranges from 8-12 download. It is obvious that they are selectively caching.

Re:Some more info

[Mezoth]

The higher your latency to a site (aka, how long does it actually take a packet to get there) the slower your bandwidth will be without some extensive TCP window tuning. Reference:

http://en.wikipedia.org/wiki/TCP_Tuning

That is why you get slower speeds the farther out you go, even assuming every link in between is not congested.

  Also, most of what you post there is explained by powerboost - very fast downloads for the first few seconds, then they throttle it back. If you ran a sustained download (go get a linux distro!) you could see what the actual sustained download speeds are after about 5 minutes.

REAL BANDWITH TEST

[logik3x]

http://miranda.ctd.anl.gov:7123/ Can't stand seeing people using speedtest.net, MOD THIS UP!

Re:REAL BANDWITH TEST

[X0563511]

And further down are links to other systems... most are working

Re:REAL BANDWITH TEST

[Kamokazi]

I think THEY just got a bandwidth test...