cancel
Showing results for 
Search instead for 
Did you mean: 

Latency highly variable but bad for simple browsing - several seconds regularly

MarkJFine
Associate Professor

Re: Latency highly variable but bad for simple browsing - several seconds regularly

Here's the thing that bothers me about off-network latency tests: You're basically testing the latency between you and their server, using their processing, which is highly dependent on several other factors.

 

To the average user that doesn't understand networking, it would almost certainly and incorrectly be interpreted as a HughesNet latency, when in fact that's only a part of it.

 

It does not include, nor does it show:
1. Any other part of the path that may be contributing (including CenturyLink and beyond).

2. any possible processing delays or congestion on TestMy's part.

 

Bottom line: The only answer they can possibly give is something more than the actual number, which is not really a good thing. As we've seen, CL can be a major contributor at a rate that's much greater than HughesNet's speed-of-light long-distance transmission latency.

 

Edit: If you're looking to assess HughesNet's internal latency, the Terminal Connectivity test in the SCC or a simple traceroute the gateway's provider can answer that. Can also compare that to a traceroute to one of the TestMy servers' IP - you can find it by doing an rDNS or simple host command.


* Disclaimer: I am a HughesNet customer and not a HughesNet employee. All of my comments are my own and do not necessarily represent HughesNet in any way.
MarkJFine
Associate Professor

Re: Latency highly variable but bad for simple browsing - several seconds regularly

Here's an example. Did a test using the LA server because it's closest to San Diego, where my gateway is.

 

The LA server is lax.testmy.net (as they say on their test) or 144.202.123.145.

 

When I use HughesNet's connectivity test, I get an average of 602mS to San Diego (563min to 650max).

 

When I use TestMy's latency test, I get an average of "1597mS to LA, 32% deviation (508.9mS)", whatever that means.

 

So there's roughly a difference of around an entire second between HN's SD gateway and TestMy's LA server.

 

The traceroute (a real mess, btw), looks like this:
Screen Shot 2019-12-19 at 1.53.42 PM.jpg

 

In sharp contrast, the actual latency to LA seems to be a lot less (by ~500mS) that what TestMy is reporting, and TestMy doesn't really tell me where the latency is. The traceroute seems to indicate it's somewhere between hop 7 and TestMy, but I have no idea where.


My only conclusion is that latency to SDO is around 600mS, latency from SDO to LA is another 500mS, and TestMy seems adding another 500mS in processing time: 600+500+500 = 1600 (or 1597mS in this case).


* Disclaimer: I am a HughesNet customer and not a HughesNet employee. All of my comments are my own and do not necessarily represent HughesNet in any way.
MrBuster
Senior

Re: Latency highly variable but bad for simple browsing - several seconds regularly

Yes, excellent points, which is why I mostly just give that screen a look over, using the drill down on the check or X to get details and scroll way down to 'RTT Diagnostic Statistics' section if I am really dying to know the ms.

 

I suppose using traceroute, then capturing a repeating ping to that step after the sat jump can give a good idea of how the number changes with time of day...maybe.

 

lax.testmy.net.png

 

>ping 2001:5b0:4600:fffa::105 -t

Pinging 2001:5b0:4600:fffa::105 with 32 bytes of data:
Reply from 2001:5b0:4600:fffa::105: time=639ms
Reply from 2001:5b0:4600:fffa::105: time=633ms
Reply from 2001:5b0:4600:fffa::105: time=619ms
Reply from 2001:5b0:4600:fffa::105: time=619ms
Reply from 2001:5b0:4600:fffa::105: time=585ms
Reply from 2001:5b0:4600:fffa::105: time=610ms
Reply from 2001:5b0:4600:fffa::105: time=585ms
Reply from 2001:5b0:4600:fffa::105: time=620ms
Reply from 2001:5b0:4600:fffa::105: time=608ms
Reply from 2001:5b0:4600:fffa::105: time=664ms
Reply from 2001:5b0:4600:fffa::105: time=600ms
Reply from 2001:5b0:4600:fffa::105: time=633ms

Ping statistics for 2001:5b0:4600:fffa::105:
Packets: Sent = 12, Received = 12, Lost = 0 (0% loss),
Approximate round trip times in milli-seconds:
Minimum = 585ms, Maximum = 664ms, Average = 617ms
Control-C
^C

 

 

I suppose it could be watched like that?

 

 

 

By the way, I think that lax.testmy.net might have been the one site in that testmy graph I made this morning where there was the big latency. I remember thinking it was funny at the time.

 

 

I am not sure how much this helps the original poster -- he was asking specifically about lag web browsing some site?  Maybe it would be good to just know which site?

 

I was drawn in just because of the mention of the CHY gateway having an issue of some sort, but maybe the issue with level3 is very random, or I move too slow to notice?

 

 

MarkJFine
Associate Professor

Re: Latency highly variable but bad for simple browsing - several seconds regularly

It seems completely random, but again, something must be driving it. Doesn't seem logical that systems just go dormant like that. Typically congestion is a good cause, and I'd put even money on it being an overload of vulnerability probing bots from China.


* Disclaimer: I am a HughesNet customer and not a HughesNet employee. All of my comments are my own and do not necessarily represent HughesNet in any way.
RandyPAlex
New Poster

Re: Latency highly variable but bad for simple browsing - several seconds regularly

The stats and data only matters to me if it helps get to the root issue of why this tremendously fast bandwidth with 1second lag feels as slow (or worse) than <100ms lag but <1 Mbps bandwith. If the metrics aren't showing it, then it seems we aren't looking at the right metrics. I had to reload this page multiple times tonight. Hughesnet,com took like a 60 seconds to load. The router diagnostics show no RTT errors. Traceroutes look pretty good. Even the latency test iisn't as bad tonight but performance is. Below are trace routes for all the Latency sites while the latency test suite ran. I tagged the two peaks. Each line is a series of consecutive TCP hits to a site. Then the next site same thing done.   I then repeated the process a second time while capturing a wireshark trace to get it all and at the tail end of the trace I hit Hughesnet.com, and Amazon.com.  At the tail end of these pics is the wireshark trace with the TCP messages sorted. Looks like lots of errors, repeats, duplicates, etc. So if something is trashing the messages in transit, no ICMP message would show it and even the TCP ping (Latency test) would only show it to some degere, right?  So now next the data sets...

 

MrBuster
Senior

Re: Latency highly variable but bad for simple browsing - several seconds regularly

Well, I thought I would run that testmy latency test to see how it differs for me in peak prime time, but it looks to me like this test really does not work well at all since the effect of the additional load of prime time really is not clearly reflected.

 

My average ping is similar(?) -- and how can the average be near 650 if there was a deviation close to 300?  If there was a ping of 950, it would take a lot of 600ms pings to pull the average down to 650, right?  I am guessing this is off or that this is stating things are choppier (busier), but that the average ping has not come up like I normally expect.  Also, in looking at the scale of sites, really nothing has changed -- almost as if the latency of a more crowded evening satellite beam was unmeasurable compared to the backbone traffic crowding that happens at any time.  Oh, and lax was not available either.

 

The times on this test seem to imply my beam is empty, unless the deviation is the only useful clue.  I think you guys were correct that this test is not too reliable at this point.

 

I estimate the load on my beam by checking pings all the time -- a choppy bunch means busy, and flat means no one is on the system without really needing to do a bandwidth test to know.  Maybe this is what it is showing and I don't know how to read this graph, but it seems easier to see a sequence of pings going all over with a rising average to realize things are getting busy.

 

testmylatency2.png

 

Looking at the choppy graph on the bandwidth test does a better job of revealing the traffic out there, but the speed does seem good for this time at night -- Did CHY get upgrades earlier this month?  It seems to be performing better than I expect for December prime time, and the last couple weeks have seemed strong.  What was that Dec. 6th morning outage?  Could that have been a quick bunch of upgrades?  Did someone spy activity in the middle of the night and assume there were big problems being fixed?

 

testmy20191219.png

MrBuster
Senior

Re: Latency highly variable but bad for simple browsing - several seconds regularly


@RandyPAlex wrote:

The stats and data only matters to me if it helps get to the root issue of why this tremendously fast bandwidth with 1second lag feels as slow (or worse) than <100ms lag but <1 Mbps bandwith. If the metrics aren't showing it, then it seems we aren't looking at the right metrics. I had to reload this page multiple times tonight. Hughesnet,com took like a 60 seconds to load. The router diagnostics show no RTT errors. Traceroutes look pretty good. Even the latency test iisn't as bad tonight but performance is. Below are trace routes for all the Latency sites while the latency test suite ran. I tagged the two peaks. Each line is a series of consecutive TCP hits to a site. Then the next site same thing done.   I then repeated the process a second time while capturing a wireshark trace to get it all and at the tail end of the trace I hit Hughesnet.com, and Amazon.com.  At the tail end of these pics is the wireshark trace with the TCP messages sorted. Looks like lots of errors, repeats, duplicates, etc. So if something is trashing the messages in transit, no ICMP message would show it and even the TCP ping (Latency test) would only show it to some degere, right?  So now next the data sets...

 


I didn't see this response -- I probably forgot to hit refresh before posting.  You mentioned you had poor speed loading the testmy page?  It seemed fast to me, GabeU/Mark any ideas?  I am seeing what looks like only mild load and great performance in the middle of the heaviest traffic season, and he seems to see something else.

 

 

Maybe try a testmy bandwidth test to see how choppy his bottom graph is compared to mine?  We are both the same gateway, so just the beams are different, and both appear to be uncongested....

GabeU
Distinguished Professor IV

Re: Latency highly variable but bad for simple browsing - several seconds regularly


@MrBuster wrote: 

Looking at the choppy graph on the bandwidth test does a better job of revealing the traffic out there, but the speed does seem good for this time at night  


This is what I've tended to rely on for evidence of congestion as well.  The choppier the heavier, though it doesn't always seem to affect my speed in the same way.  With all the other things out there that can affect speed, though, I wouldn't really expect it to.   


Ryzen 5 3400G | MSI B450M Pro-M2 MAX | 16GB Corsair Vengeance DDR4 3000 | XPG SX8200 Pro 512GB NVMe | Windows 10 Pro
RandyPAlex
New Poster

Re: Latency highly variable but bad for simple browsing - several seconds regularly

Tracert_Dallas_201912191820.PNGTracert_CO_201912191820.PNGTracert_FL_201912191820.PNGTracert_NY_201912191820.PNGTracert_SF_201912191820.PNGTracert_LAX_201912191820.PNGTracert_Toronto_201912191820.PNGTracert_uk_201912191820.PNGTracert_JP_201912191820.PNGTracert_SG_201912191820.PNGLatency_201912191829-1.pngLatency_201912191829-3.pngTracert_De_201912191820.PNGTracert_Google_201912191820.PNGTracert_Amazon_201912191820.PNGLatency_201912191829-2.pngTracert2_Dallas_201912191840.PNGTracert2_CO_201912191840.PNGTracert2_FL_201912191840.PNGTracert2_NY_201912191840.PNGTracert2_SF_201912191840.PNGTracert2_LAX_201912191840.PNGTracert2_Toronto_201912191840.PNGTracert2_UK_201912191840.PNGTracert2_De_201912191840.PNGTracert2_JP_201912191840.PNGTracert2_SG_201912191840.PNGTracert2_AU_201912191840.PNGTracert2_Cloudflare_201912191840.PNGlatency2-201912191845-1.png

RandyPAlex
New Poster

Re: Latency highly variable but bad for simple browsing - several seconds regularly

Well that was a real pain! Spent all that time since my last post just attempting to upload the pics. Had to reload so many of those that the site stopped me saying I'd loaded 100 images in the last 24 hours.  

Well tomorrow I'll upload the image of the TCP messages from the wirshark trace. Looks like lots of messages needed to be requested again and many duplicates were flagged. Seems like a lot of churn though individual speeds aren't bad??   

So I get it the latency test doesn't square with things you all see and it doesn't seem to reflect my poor performance either.

Is there a test I can run from my end that you do trust? I hope it is something on my end so that it is fixable. Just let me know what to run, please.