dslreports logo
site
 
    All Forums Hot Topics Gallery
spc

spacer




how-to block ads


Search Topic:
uniqs
1147
share rss forum feed
iansltx

join:2007-02-19
Austin, TX
kudos:2
Reviews:
·Time Warner Cable
·Verizon Online DSL

They should partner with someone on this for servers

Not sure where the Ookla server is, but unfortunately M-Lab appears to be on a not-so-fast network. If that's Google's, that sort of explains it (GOOG isn't known for excellent connectivity with everyone under the sun). On the Ookla side my test originated from their DFW server and gave me very similar results to what I'd get on Speedtest/PingTest; Ookla has good connectivity there.

Of course the problem here is that the test isn't long enough to negate PowerBoost...see below for my most recent test results (keep in mind that I'm on Comcast 12/2):

Ookla
Download: 20826 kbps
Upload: 3451 kbps
Latency: 34 ms
Jitter: 3 ms (cable connection + wireless to my computer)

M-Lab
Download: 8502 kbps
Upload: 3520 kbps
Latency: 56 ms
Jitter: 39 ms (looks like there's a problem with the speedtest server)

So one test gets things totally wrong, and the other test includes PowerBoost.

It's not like there aren't tools out there to accurately measure after-boost speeds:

»www.cc.gatech.edu/~partha/diffpr···obe.html

That gives me the correct speed info...

FFH5
Premium
join:2002-03-03
Tavistock NJ
kudos:5

Re: They should partner with someone on this for servers

said by iansltx:

It's not like there aren't tools out there to accurately measure after-boost speeds:
That isn't the point. This is:
»FCC using to build own broadband maps
iansltx

join:2007-02-19
Austin, TX
kudos:2

Re: They should partner with someone on this for servers

I'm aware of this, however if you have a speed test it might as well be accurate.

FFH5
Premium
join:2002-03-03
Tavistock NJ
kudos:5

Re: They should partner with someone on this for servers

said by iansltx:

I'm aware of this, however if you have a speed test it might as well be accurate.
They are as accurate as the underlying vendors(OOKLA& MLAB) tests that are providing the service to the FCC. More accurate tests(that handle dealing with powerboost) take too long and consume too many resources for the numbers of people who use those sites.

If you want a test that handles distortions caused by powerboost, use this one:
»www.measurementlab.net/measureme···iffprobe
iansltx

join:2007-02-19
Austin, TX
kudos:2
Reviews:
·Time Warner Cable
·Verizon Online DSL

Re: They should partner with someone on this for servers

Uhh, I actually mentioned that test in another post. That exact one, in fact.

For what it's worth, it wouldn't take a whole lot to get accurate results on speed tests money-wise; I'm guessing Cogent and/or Hurricane Electric would be okay with providing a few gigabits of bandwidth for a nominal fee (Cogent is pro-net neutrality btw), and a few midrange quad-core servers on gigabit pipes, limited to twenty simultaneous tests, would probably do well for this sort of thing...

funchords
Hello
Premium,MVM
join:2001-03-11
Yarmouth Port, MA
kudos:6
said by iansltx:

M-Lab
Download: 8502 kbps
Upload: 3520 kbps
Latency: 56 ms
Jitter: 39 ms (looks like there's a problem with the speedtest server)
If the M-Lab test is not sane, try this:
»ndt.iupui.donar.measurement-lab.org:7123/

It will give you more information than Ookla attempts to detect. It also discloses what server you're talking to. (Note: this bypasses the FCC, but the test run is stored by MeasurementLab and becomes freely available to researchers).
--
Robb Topolski -= funchords.com =- District of Columbia -- KJ7RL
Tweet! Tweet! -- »twitter.com/funchords
iansltx

join:2007-02-19
Austin, TX
kudos:2

Re: They should partner with someone on this for servers

PowerBoost affects the speed of the link still, but at least it's pretty much maxing out the connection rather than giving me a bogus result...

CW

@t-mobile.com
The Ookla and M-Lab uses different test algorithm to calculate throughput. M-Lab reflects user experience throughput=total bits (bytes)/time used
Ookla checks the instant throughput at rate of 30 samples per second, and average the upper half samples, which reflects 70-80 percentile of peak throughput