US Cable Support > Mediacom > Attention Speedtest Users, please read.

Attention Speedtest Users, please read.

Yes, I too use testmy.net often and agree, it seems to be much more realistic on what you're actually getting since it's browser based, not flash.



This is about normal, not even close to 50Mbps d/l. I saw 47Mbps once or twice maybe over the past 9 months.

reply to bulletsam
All Flash-based tests are all but accurate. They look good, and that is about it.

reply to bulletsam
You also have to take into account that mediacom uses boosting which effects speedtests accuracy.

Also the number and kind of plugins used on browsers affects things as well.

Speedtest's are a great tool in gathering information. Unfortunately you can't use that information to really troubleshoot the issue. While there are many different speedtest sites out there I have noticed all of them can give wonky results at times and although they are great tools, cannot be trusted 100% of the time.

reply to bulletsam

reply to bulletsam
only use testmy.net - everything else is ISP cached crap

make sure to default the correct regional server for your area.

sometimes i test from NY:TX when I should be testing from NY:DC

reply to bulletsam
I never noticed it until you posted about it, hehe. I did several tests in Waterfox 18.0.1/Flash 11.5 and IE10/Flash 11.3, on Comcast. IE10 got 5-15 ms, Waterfox got 25 ms consistently. Doing traceroutes to near the test server I picked (San Jose) were closer to 20 ms, but I wouldn't rely on numbers that precise with ICMP.
--
KI6RIT

reply to bulletsam
I've done a lot of testing with speedtest.net and testmy, and from what I've seen, speedtest.net is far more accurate. Contrary to popular belief, testmy has big issues and is very inaccurate. If you want more accurate results and distrust speedtest.net for some reason, try testing to cachefly (»cachefly.cachefly.net/100mb.bin), or via ShaperProbe. If you're having issues maxing out your line with single-threaded downloads, install lftp (or any similar threaded download manager).

well JTR, I'll agree to disagree.

speedtest certainly has the greatest number of test servers worldwide, but it's results are the same ISP cache inflated near city nonsense you get from most test sights including DSLR's. testmy is accurate as you can specify whatever size down/upload you want to work with within a given region, so it better simulates real world applications. how many web content or game hosts are within the 50 miles speedtest is using to get to you - very few to be sure.

i always test size 40mb down and 2mb up as that is double the 20/1 speed rating my ISP is supposed to be selling me, and i get consistent test result every time that best reflect prime time and after hours congestion. longer distance tests also help reveal when your ISP is giving you poor routing.

the absolute proof to me is when i actually use a download client or uTorrent and see my top speed never actually eclipsing 2.3MB/s - according to speedtest's bogus results, i should be seeing nearly 3.3MB/s. speedtest is such marketing garbage they can't even label their panel right... it should be Mbps not Mb/s - what fools, there is no such abbreviation standard - they just made that shit up.

»web.forret.com/tools/bandwidth.asp



useless rubbish:

reply to JTR

reply to Packeteers
Let's take a look at your claims. The first one? You claim that speedtest servers are cached by ISPs. Quite an interesting claim. Let's take a look at speedtest.net's public list of test servers.

»speedtest.net/speedtest-servers.php

2045 servers in that list. Last time I checked, each server has ten files on it, in the sizes of 240KB, 494KB, 1.1MB, 1.9MB, 4.3MB, 7.5MB, 12MB, 17MB, 23MB, and 30MB. Totaled up, that's around 97MB. If your claims were true, and every ISP cached the entirety of these test servers, that'd be around 200GB of stuff. Quite a hefty amount of data, but by no means too large to cache.

HOWEVER, your claim has a extremely large hole. IF the ISPs cached the files (as you claim), it would be extremely easy to see that the file was coming from the ISP, not from the speedtest server.

To test this theory, I used speedtest.net to test to the Dexter, MO server (ID 2648). »speedtest.net/speedtest-servers.php has a URL of »speedtest.4bxn.net/speedtest/upload.php listed for this test server. speedtest.4bxn.net resolves to 12.234.219.238. In order to determine if Mediacom caches the file from that server, I ran netstat -n during the test. The results revealed that my computer connected to 12.234.219.238 on port 80 over six TCP sockets during the test. I don't have iftop installed, so I cannot directly confirm the volume of data being transferred from that IP, but it seems to match up.

A simple traceroute to 12.234.219.238 reveals nothing unusual. All hops look normal. There's no indication of it routing in an unusual manner. Sure, it's possible to hijack the HTTP stream, but it'd be a huge pain in the ass to do — impractical to attempt at a large scale like this.

SO, your claim of ISPs deliberately caching speedtest.net results is shaky at best, and a blatant lie at worst.

Onto the next claim! Your claim is testmy is more accurate. But where is your proof?

I'm going to run some tests. Lets see.

This test is only intended to measure download speed.

Testmy (Dallas, TX): 22Mbps
Speedtest.net (Dexter, MO server): 21Mbps
Speedtest.net (St. Louis, MO server): 23Mbps
Speedtest.net (Paducah, KY server): 28Mbps
Speedtest.net (Chicago, IL server): 27Mbps
Speedtest.net (Dallas, TX server): 27Mbps

Hmm. I'm noticing a few things. First of all, the Paducah and Chicago servers seem to be using speed boost, while the St. Louis and Dexter servers are not. Testmy doesn't seem to be using it either.

You may notice something interesting about testmy.net. In my case, it tested to a server based in Dallas, TX. A speedtest.net test to a Dallas, TX server showed that I was able to attain speeds of 27Mbps.

HOWEVER, there is a crucial difference here. Speedtest.net is multi-threaded, testmy is single-threaded. Whats more, routing may play an important part in this. This calls for further investigation.

Testmy's IP seems to be owned by softlayer. Let's head over to their looking glass and find some downloads to test with. I'll test it multiple times to emulate speedtest.net, testmy.net, a full 100MB file, and more.

»speedtest.dal05.softlayer.com/do···t100.zip (single-threaded, 15MB downloaded (emulates testmy)) :: 20Mbps (2Mbps less than testmy)
»speedtest.dal05.softlayer.com/do···t100.zip (single-threaded, 30MB downloaded (emulates speedtest.net without the threading)) :: 26Mbps (1Mbps less than speedtest.net, despite being single threaded)
»speedtest.dal05.softlayer.com/do···t100.zip (single-threaded, full 100MB downloaded) :: 25Mbps
»speedtest.dal05.softlayer.com/do···t100.zip (5 threads, 30MB downloaded (emulates speedtest.net with the threading)) :: 25Mbps (2Mbps less than speedtest.net)

Hmm. Looks like speedtest.net is only marginally faster than either singlethreaded or (extremely crude) multithreaded transfers of the same size. Testmy lags behind. If your goal is to measure connection speed, you want to saturate the connection with multiple threads, right? Why is testmy considered more accurate when it's slower and its results cannot be reproduced with curl or lftp? You can argue that speedtest.net's practice of using multiple download threads is misleading, but the results it produces are at best 1Mbps above what can be attained with a single-threaded HTTP download from curl, and the goal is to test max connection speed, not post-speedboost speed.

Of course, I used softlayer's test, not speedtest.net's test. So let's do two more raw HTTP tests with speedtest.net's softlayer file, just to eliminate any possible issues with different test servers.

»speedtest.dal05.softlayer.com/mi···4000.jpg (single-threaded) :: 27Mbps (same as speedtest.net)
»speedtest.dal05.softlayer.com/mi···4000.jpg (5 threads) :: 25Mbps (2Mbps less than speedtest.net)
»speedtest.dal05.softlayer.com/mi···4000.jpg (10 threads) :: 25Mbps (2Mbps less than speedtest.net)

Looks like speedtest.net is pretty accurate. Not much skewing going on here from threading.

I wonder what cachefly will show?

»cachefly.cachefly.net/100mb.bin (single-threaded :: 26Mbps
»cachefly.cachefly.net/100mb.bin (2 threads) :: 27Mbps
»cachefly.cachefly.net/100mb.bin (5 threads) :: 27Mbps

I wonder what my chicago-based VPS will show. It's not exactly a premium VPS (network performance can be terrible at times), there's no way this is being cached by Mediacom, it should be a good indicator of what my speeds are, right?

(single-threaded) :: 27Mbps
(2 threads) :: 26Mbps
(5 threads) :: 26Mbps

I'm not seeing much of an advantage with multiple threads, it all seems to be within normal margins of error.

So, in conclusion, testmy is only "more accurate" if you want an arbitrary number that may or may not reflect your post-boost speeds. ShaperProbe does a much better job of doing that. As I have demonstrated here, speedtest.net is NOT being cached, and the speeds on it seem to accurately reflect what you see with single and multi threaded tests ran via curl, lftp, etc.

Thus, there is no evidence whatsoever that speedtest.net's speeds are artificially inflated past what you will see with realistic usage. The only flaw is that speedtest.net's results may vary based on server load and routing, but this is an unavoidable part of how the internet backbone operates. I suggest testing to multiple servers to get an accurate picture of your speeds.

Tl;Dr: Speedtest.net accurately reflects your current speeds, ShaperProbe accurately reflects your actual speeds, testmy is unreliable and may or may not accurately reflect your speeds.

Edit: Also, testmy is very inaccurate at high speeds. I used to rent a server with access to 1Gbps of bandwidth - speedtest.net (remote and local servers), cachefly, softlayer, torrents, etc all reported around 800Mbps-1Gbps, while testmy reported speeds of around 200Mbps (despite easily being able to max the line). There was NO speedboost WHATSOEVER in this particular case — testmy was simply incapable of providing accurate results.

ISPs gladly pay speedtest and host other such ridiculous brief localized test sites to make ISP's look good, so it's no wonder testmy does not bother or can afford supporting speeds well above 100mbps - speeds maybe one in 10,000 of us experience.

»www.timewarnercable.com/en/resid···est.html

above is an example how little ISP's care to accurately test anything. I got 31Mbps on my 20Mbps rated line at over 3.9MB/s transfer rate? do these corporate stooges really think we are all that gullible? I've downloaded Windows install discs directly from MicroSoft (with or without a download manager), torrent'ed Ubuntu from 1GB University Hosts through my ISP (no VPN) and I never got anywhere above 2.3MB/s on my line even after hours - so I'm sorry, but here on planet Earth, this kinda test is total horse manure



there is a very simple explanation why ISP's pull this kind of crap on all of us - it's to save on support costs. if your 20/1 ISP is giving you 10/1 today and you call support, they point you to that test above and you'll test out at an inflated 20/1 so the ISP can say - see? everything is fine, now go back to your knitting... give me a break.

»www.broadband.gov/qualitytest/about/

last year the FCC even sponsored the above OOKLA speedtest submission system to document how well ISP's were delivering on their rated speeds. both the FCC and ISP's undoubtedly colluded to make sure we all got such inflated readings, as to make them both looked like a job well done - seriously?