dslreports logo
 
    All Forums Hot Topics Gallery
spc
Search similar:


uniqs
3556

justin
..needs sleep
Mod
join:1999-05-28
2031
Billion BiPAC 7800N
Apple AirPort Extreme (2011)

justin

Mod

speed tests over satellite

Ages ago I converted the java speed test for Hughes to use on their site under their label. Now I can't quite remember how it should be different from any old speed test.

Does Satellite still compress uploads if they can be compressed?

Anyway I'm looking for a few troopers to tell me whether

»/speedtest

is working as well as can be expected, for the "Satellite" button, in particular the upload phase and graph.

thanks.

gwalk
Premium Member
join:2005-07-27
West Mich.

gwalk

Premium Member

Justin,

As the differences in the Hughes "house brand" speed test and others ... it is best described by this old but still valid comment by Hughes Engineer Patrick Fisher:

Patrick Fisher (Employee)

Also, a little more information on why we have our own test. (And no, it's not so we can doctor results).

Speed Test sites have many ways at their disposal to measure your speed, and they all do things differently. Some use Adobe Flash to measure speeds, some use Java, and some use your browser and JavaScript. Sometimes they use multiple connections, and maybe to multiple servers. Or maybe just a single connection. Some try to detect your location, which is tricky because as a satellite network, the address that matters is the location of your gateway, not your own location or where the IP address is registered. So they might pick a server in Kansas, or you might pick a server in New York where you live, but your gateway is in Washington state, thousands of miles away. Then, just to complicate things, we have a number of acceleration capabilities which try to improve performance over satellite by compensating for latency.

The end result of all of these variables is that we have no idea what they are actually measuring. A Speed Test might think it is giving you more accurate results by using weird tricks to get multiple measurements at once and combine them, but maybe our acceleration software interferes with that trick and makes it look better or worse than it is. Or maybe the speed test you pick has congestion on the internet link between them and us. Or maybe they pick a server that's thousands of miles from our gateway, so you have to deal with a bunch of extra internet latency. When we have all of these questions (and when we have no control or even information from the speed test provider about how it works), then we don't really get any useful information from external speed tests that we can use to troubleshoot. Finally, the one variable that they do show - uncontrolled internet links between us and them - is both rarely a problem, and not something you or Hughes is able to fix.

Our speed test is a simple test, which is designed to do us a couple favors:

1) We use Java, so that we can completely bypass your browser and any extensions or plugins, like Adobe Flash, that might introduce other variables. We don't have to question whether Adobe is doing weird things or your browser is crawling because of other extensions.

2) We log the tests, both so that we can see the tests you as an individual have run, and so that we can run reports on all of the speed tests for users in a given area, on a specific gateway, etc. This helps us identify performance trends versus performance problems that might be caused by a single user's configuration or LAN. If everyone else on your gateway or in your beam is getting good speed test results, but you are not, we can zero in on the things that are unique.

3) We control, manage, and monitor the links between gateways and speed test servers, so we can eliminate external factors like internet bandwidth as a culprit.

4) We can skip certain satellite optimizations as described above.

We do actually go out of your gateway on the same internet connection that your real traffic uses, to get to our speed test servers. So we aren't cheating by under-sizing our internet pipes and showing you just the satellite link, either.

Hopefully this helps shed some light on why we encourage users to use our own speed test service.

..................

One unfortunate issue with TestMy.net (since they really do have a very good, simple and accurate test methodology) is that our compression software is particularly good at compressing their data.

Much of the web traffic that we send toward your terminal gets compressed by our acceleration software. What that means is that if a web page has a 1MB JPEG image, we might compress it and only send 500KB over the satellite link before it gets decompressed on the other side by your terminal. If your connection is 2Mbps, we can effectively send you that 1MB image at a speed of 4Mbps. (As an added bonus, only the smaller size gets counted against your download allowance - we always pass the benefits of our acceleration on to users)

TestMy.net's data gets compressed by this, so they send you a (for example) 2MB file, we compress it on the fly to 500KB, and they get 4 times the actual raw speed of the link.

This makes their test accurate for seeing the effective throughput for compressible data, but not necessarily for something like a bulk file transfer. The Hughes test shows the actual raw speed of your connection, skipping that web acceleration.

Justin, hopefully that will give you some insight as to the Hughes "architecture".

justin
..needs sleep
Mod
join:1999-05-28
2031

justin

Mod

Hmm as I guessed mainly it is about compression.

Luckily, the upload file I use is incompressible, so I'd be interested if anyone can do A-B runs of the Hughes recommended test and ours at /speedtest ..

gwalk
Premium Member
join:2005-07-27
West Mich.

gwalk

Premium Member

Click for full size
Here is one of yours I just ran ...
I question the ping, 453 ms is imposable on satellite ... 580 min.

I will post a comparison to the Hughes and one other speed test that is an accepted alternative in a moment.

justin
..needs sleep
Mod
join:1999-05-28
2031

justin

Mod

Can you post the link to the page you get when you press the share button?

gwalk
Premium Member
join:2005-07-27
West Mich.

gwalk

Premium Member

Click for full size
this one is considered a acceptable alternative on the Hughes Community:
»www.att.com/speedtest/
gwalk

gwalk

Premium Member


justin
..needs sleep
Mod
join:1999-05-28
2031
Billion BiPAC 7800N
Apple AirPort Extreme (2011)

justin to gwalk

Mod

to gwalk
I got the link and looked in the log.

1. The latency results in the log show it took 654ms to do a minimal fetch from »54.153.0.97 .. now normally to get ICMP, you can divide that by two because a web server transaction requires at least two round trips not one round trip. So what is your ICMP ping over satellite to 205.251.229.106 (last hop before 54.153.0.97) ?

2. the IP I see in the log that ran the test, 69.35.X.X is that a proxy located at Hughes? If I ping it, I'm pinging Hughes, not you at the end of the link?

3. the download speed was still rising where normally (other technologies) it has long since flattened so it would score a higher download if I wait longer. About how long in seconds is the download phase of the official test or the AT&T test?

4. the upload is reported as 3 megabit which is about twice what the AT&T test gives. What is your opinion of actual satellite upload speed on a good day? 1.4mbit ? or it could be 3 megabit? before I look more carefully at the upload payload I want to check if 3 megabit is real.

thanks!

gwalk
Premium Member
join:2005-07-27
West Mich.

gwalk

Premium Member

Click for full size
Here is the in house Hughes speed test:
their protocol is to run a series of 5 tests back to back. The Hughes test does not factor in ping. their is another test for that.

justin
..needs sleep
Mod
join:1999-05-28
2031

justin

Mod

Ok so I'll look at the upload compression possibilities again.

if you get back to me on the other questions, I'm sure our test can be tweaked to give as correct an answer as any, probably better than most.

gwalk
Premium Member
join:2005-07-27
West Mich.

gwalk

Premium Member

Click for full size
 
The ping you are getting is from my "gateway".

I am located in west Michigan, my "gateway" is located in Flagstaff AZ.

To give an idea of lay-out, Gen4 EchoStar 17 has 60 "spotbeams", users must be within the foot print of a spot beam.

I am in beam 13, serviced by gateway 6.
There are however 17 "gateways", these are the ground based facilities where a users signal is "handled" Earthside"

The hughes "ping test is here:

justin
..needs sleep
Mod
join:1999-05-28
2031
Billion BiPAC 7800N
Apple AirPort Extreme (2011)

justin

Mod

No I'm not getting the ping from a gateway .. the way it works ping is calculated on your end, by your browser, it measures the time to do a complete fetch.

BUT, it divides it by two to get "ICMP" equivalent and shows it, which is why it looks low to you.

It looks like it should not be dividing by two, in which case the reported ping would be, at best, what I wrote above. 654ms, which looks bang on the money.

BTW I have tweaked /speedtest (for satellite button) to extend the download period, and to change the payload that is uploaded. Please if you could re-run? ... I'm looking for a better download speed with a longer duration, and hoping the upload will be measured correctly. The ping time will still be low because it is still dividing by two.

gwalk
Premium Member
join:2005-07-27
West Mich.

gwalk

Premium Member

Click for full size
Justin, is this what you need ?
You never will be able to ping my terminal directly.
This is why so many users are dismayed to find that security cameras and such will not work with a Hughes system without having a "static IP".
There are no more static IP available for residential service as nearly all of the Hughes owned IPv4 public addressable IP are gone.
There are some in Business grade accounts.

My "plan" rating is 15 Mbps down and 2 Mbps up.
Typically as Hughes user will not see more than 1/2 of plan rated upload speed and that has been true for me.

justin
..needs sleep
Mod
join:1999-05-28
2031
Billion BiPAC 7800N
Apple AirPort Extreme (2011)

justin

Mod

yes that is exactly right. thanks.
It confirms what I need to know.

Now the last thing, please re-run the test any time from 10 minutes from now?

edit: or any time from now.
I've confirmed the download lasts for 30 seconds now for Satellite
and the upload is done differently. So it will be interesting what results you get..

gwalk
Premium Member
join:2005-07-27
West Mich.

gwalk

Premium Member


justin
..needs sleep
Mod
join:1999-05-28
2031
Billion BiPAC 7800N
Apple AirPort Extreme (2011)

justin

Mod

sorry to do this to you, you just caught the end of the wrong version.

try now if you have the time.
I changed the number of streams to 1 as well. It is pointless to do parallel streams over satellite.
(shift-reload the /speedtest page just to be sure you're not using the stale script).

gwalk
Premium Member
join:2005-07-27
West Mich.

gwalk

Premium Member

Click for full size
upload looks like it still reads too high:

»[Satellite Speed test: 15.37/1.53 1008 ms]

justin
..needs sleep
Mod
join:1999-05-28
2031
Billion BiPAC 7800N
Apple AirPort Extreme (2011)

justin

Mod

No that is correct. It isn't the graph that is important it is the answer.. 1.53

right?

overall I think this is now about right.. even the ping (which if anything is a bit high now, because it is an average).

The best ping was 649ms
The worst was 1922ms

it shows the average ping of the "closest" server it can find and because of that variability, the average is high.
justin

justin to gwalk

Mod

to gwalk
And thanks for your patience.

Overall pretty painless just needed a bit of adjusting to take into account the, er, unique features of Satellite.

gwalk
Premium Member
join:2005-07-27
West Mich.

gwalk to justin

Premium Member

to justin
I guess it may not be too different from my history.
I seldom get above 1200 Kbps on the Hughes test but who is to say that that "yardstick" is spot on.

justin
..needs sleep
Mod
join:1999-05-28
2031

justin

Mod

if you have a windows PC you can probably run a free utility, or I think there is one in there somewhere, to watch your ethernet interface and graph the outbound bytes/second, that will get you proof or otherwise of the upload speed?

gwalk
Premium Member
join:2005-07-27
West Mich.

1 edit

gwalk

Premium Member

I am soon to replace my router with one that can run Merlin firmware and track usage by device IP.
That is a big issue with Hughesnet and its those "pesky" wireless devices that are hard to tame on a metered connection.
One thing with Hughes is that the "pipes" are so small that if you have multiple devices sharing that connection the "speed per device" can really appear to tank.

james1979
Premium Member
join:2012-10-09
Quinault, WA

james1979 to gwalk

Premium Member

to gwalk
said by gwalk:

Justin,

As the differences in the Hughes "house brand" speed test and others ... it is best described by this old but still valid comment by Hughes Engineer Patrick Fisher

I never noticed any discernible difference between HN's speed test and testmy.net when I was a Gen4 customer. I still don't notice much difference between the two tests. Despite this humorous drought rain which likely affected the speed tests, it's obvious why I canceled the $70/month service and kept the $50/month service:




I'm not trying to start an off topic "speed war" discussion, but I did recently post some testmy.net results on the "other side". Both speed tests are about the same for me. Any other speed tests that I have tried (including BBR's) seem to, uh, fail the test.
quote:
if a web page has a 1MB JPEG image, we might compress it and only send 500KB over the satellite link before it gets decompressed on the other side by your terminal.

In the truest sense of the word, it would be incredible to losslessly compress/decompress a JPEG by 50%.

justin
..needs sleep
Mod
join:1999-05-28
2031
Billion BiPAC 7800N
Apple AirPort Extreme (2011)

1 edit

justin

Mod

said by james1979:

said by gwalk:

Justin,

As the differences in the Hughes "house brand" speed test and others ... it is best described by this old but still valid comment by Hughes Engineer Patrick Fisher

I never noticed any discernible difference between HN's speed test and testmy.net when I was a Gen4 customer. I still don't notice much difference between the two tests. Despite this humorous drought rain which likely affected the speed tests, it's obvious why I canceled the $70/month service and kept the $50/month service:

I'm not trying to start an off topic "speed war" discussion, but I did recently post some testmy.net results on the "other side". Both speed tests are about the same for me. Any other speed tests that I have tried (including BBR's) seem to, uh, fail the test.
quote:
if a web page has a 1MB JPEG image, we might compress it and only send 500KB over the satellite link before it gets decompressed on the other side by your terminal.

In the truest sense of the word, it would be incredible to losslessly compress/decompress a JPEG by 50%.

when did you last run /speedtest
if in the last day why exactly do you think it "fails the test"?

edit: i see your result was before the feedback I got in this topic.
Your result looks right except for the wrong upload and low ping, both of which are now fixed. Try it again see if you dont feel the test doesnt fail the test.

james1979
Premium Member
join:2012-10-09
Quinault, WA

james1979

Premium Member

said by justin:

when did you last run /speedtest
if in the last day why exactly do you think it "fails the test"?

edit: i see your result was before the feedback I got in this topic.
Your result looks right except for the wrong upload and low ping, both of which are now fixed. Try it again see if you dont feel the test doesnt fail the test.

Justin,

When I found your topic, it was "perhaps past my bedtime", and I did not understand the context of your discussion with Gwalk. I did not follow through on the link in your first post, and I also did not infer that you are developing / working on BBR's speed test. (Which is apparently relatively new on BBR?) As such, my little "fails the test" joke would seem rude to you. I sincerely apologize for seeming rude and being unhelpful to you.

The first time that I ran /speedtest, the ping was obviously off along with the upload speed. After you replied to my previous post, I understood this discussion. I then ran /speedtest. The reported ping time seemed reasonable, yet the upload speed seemed exaggerated. I decided to investigate the matter after sleep.

A few hours earlier today I ran /speedtest twice on my Exede-12 system, and both results were very similar. Here's the first one:




The ping response seemed a bit high, and the upload speed seemed exaggerated again. Both times that I ran /speedtest, my upload speed was closing in on 5 Mb/s, and then there was a slight drop-off near the end of the test (which is visible on the graph). I just ran /speedtest again and received similar results along with the slight drop-off near the end of the upload test.

Next I ran the AT&T speed test (»www.att.com/speedtest/) which Gwalk shared earlier. I received this result:




Since I have never received an upload much above 3 Mb/s from either testmy.net or the HN speed test, I concluded that the AT&T test was also exaggerating my upload speed.

Then I decided to try something different and use the SCP program to copy a file containing 37,172,500 bytes of data compressed with bzip2 to a Linux box located in Texas. To my surprise, scp reported the copy completed in 56 seconds (5.31 Mb/s). I tried the scp test again and got the same result. I repeated the test yet again and scp reported the same transfer speed. To rule out a bug in scp, I downloaded a stopwatch app which rounds to 1/10ths of a second and manually measured the time that it took to upload copy the file. It took around 56 seconds! Finally, I twice copied a WAV file containing 132,808,076 bytes. My upload speed was 5.33 Mb/s.

"Oh, so I got it backwards. The HN speed test and testmy.net are faulty in measuring upload speeds." So I went to testmy.net and performed an upload speed test with with 25 MB of data (much larger than I have tried in the past). And ...



Note that scp bypasses compression by any ISP (or else scp has been compromised and isn't secure), so it did not matter that I used .bz2 or .wav files. I could (and just did) use scp to upload a file containing 30 million identical bytes at a transfer rate of 5.38 Mb/s which confirms that Exede is not trying to compress encrypted data.

No as for the ping time of /speedtest, it seems to be a bit inflated. That is not the case however. A snippet of a manual ping to google.com indicates that Exede has introduced some latency variation since I last checked my latency:


64 bytes from 173.194.115.34: icmp_seq=48 ttl=54 time=1004.748 ms
64 bytes from 173.194.115.34: icmp_seq=49 ttl=54 time=733.667 ms
64 bytes from 173.194.115.34: icmp_seq=50 ttl=54 time=961.761 ms
64 bytes from 173.194.115.34: icmp_seq=51 ttl=54 time=698.748 ms
64 bytes from 173.194.115.34: icmp_seq=52 ttl=54 time=655.522 ms
64 bytes from 173.194.115.34: icmp_seq=53 ttl=54 time=655.990 ms
64 bytes from 173.194.115.34: icmp_seq=54 ttl=54 time=953.574 ms
64 bytes from 173.194.115.34: icmp_seq=55 ttl=54 time=890.384 ms


So in summary, »/speedtest works rather well! The small drop-off near the end of the upload tests occurs repeatedly for me, and it is skewing the final test result down:




Oh, my explanation as to why scp is revealing the technical limits of my uplink (which are higher than any other speed test that I have used) is that the satellite ISP's compression algorithms are actually slowing down transfers of data that can't be compressed.

Thanks for your work!

justin
..needs sleep
Mod
join:1999-05-28
2031
Billion BiPAC 7800N
Apple AirPort Extreme (2011)

justin

Mod

said by james1979:

So in summary, »/speedtest works rather well! The small drop-off near the end of the upload tests occurs repeatedly for me, and it is skewing the final test result down:

Thanks for so comprehensively investigating things. I'm glad it is working and everything makes sense.

The upload measurement is done as a MEDIAN (not average) of the stream of instantaneous upload measurements that are graphed, and rather uniquely in your satellite result, vs other technologies, there is a long slow ramp up of speed at the start, I think this is pulling the median "final result" down a little (but not a lot, because it is a median, but that is the flat part at the end).

the solution I think is to ignore more of the wind-up part, and concentrate on a median of the latter parts.

Or alternatively and in the case of satellite only, report the highest peak found as the result, rather than the median. If the highest, then it would report very close to 5.0

Which do you feel is better?

gwalk
Premium Member
join:2005-07-27
West Mich.

gwalk to james1979

Premium Member

to james1979
That is interesting James1979, thanks for sharing.
gwalk

gwalk to justin

Premium Member

to justin
Click for full size
latest test, seemed pretty good.
»[Satellite Speed test: 15.56/1.85 731 ms]

c0rr0sive88
@direcway.com

c0rr0sive88

Anon

»i.gyazo.com/ca2bd73da430 ··· e78b.png

I can only say that my latency sure isn't 1800ms to anyone unless it's to euro.

justin
..needs sleep
Mod
join:1999-05-28
2031
Billion BiPAC 7800N
Apple AirPort Extreme (2011)

justin

Mod

said by c0rr0sive88 :

I can only say that my latency sure isn't 1800ms to anyone unless it's to euro.

For a few hours I lost the correct calc of latency because I made some changes and satellite is different to terrestrial.
It should be right now.