|
Is this throttling?Just moved to the NE from Charlotte and had to end up with Comcast. I got the highest speed tier, up and down thinking i wouldn't have any issues I am trying to pull down a large file from utorrent and saw an interesting pattern. Is this trottling by Comcast? |
|
|
Comcast does not throttle, so no. |
|
|
bradyrYCCD - Network Operations Premium Member join:2008-10-27 Sonora, CA (Software) pfSense Netgear CM1200 Ubiquiti UAP-AC-HD
|
to Crispyfertoo
yanking in a torrent might not be the best platform to base a speedtest on. that graph could be powerboost (though powerboost shouldn't sawtooth like that if it's a constant download.. it should start high then ramp down and be solid), or it could be comcast interfering with your torrent download.
you need to download some big files, like some linux iso's or something and see how the download goes. |
|
|
to Crispyfertoo
|
|
bradyrYCCD - Network Operations Premium Member join:2008-10-27 Sonora, CA (Software) pfSense Netgear CM1200 Ubiquiti UAP-AC-HD
|
bradyr
Premium Member
2012-Oct-8 10:45 pm
man i just tried that testmy.net "speedtest" and it was totally inaccurate for my comcast connection. the graph was all over the place and only around 60% of the speeds that i normally pull day in/day out.
I'm on blast, and when i scp or rsync multi-gigabyte files from the office to my home computers, i average out at 3MB/s (24Mbps) all day long (powerboost gets me at 36Mbps or 4.5MB/s for the first 10-15 megs of the download). no dips, no fluctuations for the duration of the download, unless i'm multitasking on the net.
i guess what i'm getting at, is if i used that testmy.net as an indicator of the health of my connection, i would be going off of completely incorrect data. |
|
Motorola MG8725 Asus RT-N66
|
to bradyr
Sometimes with me I see a lot of up and down download speeds depending on what I'm downloading. Sometimes if I download from say EA I constantly on Performance go from 25 Mbps(Couple seconds) to 16(Couple seconds) then dips a little down to 14 or so(Very few seconds) then boosts back up to 25 and repeats that cycle. |
|
|
to bradyr
They're always accurate for me. That speedtest was also for the east coast, not west. You need to be aware of your PowerBoost though and then watch NetMeter to see your actual speed after PowerBoost falls off. |
|
bradyrYCCD - Network Operations Premium Member join:2008-10-27 Sonora, CA (Software) pfSense Netgear CM1200 Ubiquiti UAP-AC-HD
|
bradyr
Premium Member
2012-Oct-8 11:23 pm
yeah I didnt check to see where i was connecting to with your link.. my bad there.
I think we can all agree that the bottom line is if the connection is healthy, factoring in powerboost, a typical throughput graph (assuming you're testing against a non-overloaded server and you're "dedicating" your connection to the test.. i.e. telling everyone else to stop doing stuff on the net), your graph should start high, fall off some as powerboost stops, then it should be damn near a rock solid straight line for the duration of the download.
i'll try to post up some TCP stream graphs from wireshark, against some hefty sized downloads (via ftp or sftp), that shows this.
I used to do tcp stream graphs, and look at them all the time when i had dsl and i was on exhausted at&t redbacks/BRAS. you would see the graph go ape-shit (up and down up and down), during prime-time hours. not that clean looking saw-tooth pattern (that looks like throttling, kinda-sorta) and certainly not a smooth straight line. |
|
|
to Crispyfertoo
I started seeing the same pattern (extreme sawtooth pattern) downloading NNTP (Usenet) files about a week ago. All other speed tests on other types of files give me max or near max bandwidth download speeds... I have tested with 30 download threads using easynews and giganews... for about a week now, consistant pattern... peaks at max bandwidth, then down to zero, then back up again.. I have bypassed my router (direct cable modem connection) and no joy.. same results. Also tried using SSL NNTP connection - same pattern. I have been downloading NNTP using Comcast for many years.. never had this issue before. Say what you want, I am convinced that Comcast is doing some kind of traffic shaping/throttling. BTW, what I am seeing is absolutely NOT a powerboost feature. |
|
JohnInSJ Premium Member join:2003-09-22 Aptos, CA 1 edit |
JohnInSJ
Premium Member
2012-Oct-9 8:15 am
double post. |
|
JohnInSJ |
to Crispyfertoo
said by inMaryland :Say what you want, I am convinced that Comcast is doing some kind of traffic shaping/throttling. BTW, what I am seeing is absolutely NOT a powerboost feature. Comcast does do packet de-prioritization. If your node is saturated, and you have maxed out your line for X minutes, then your traffic is marked lower priority for Y minuets. Which means you'll get less of the fully saturated node's capacity. See » customer.comcast.com/Pag ··· nagementSpecifically » customer.comcast.com/Pag ··· echnique under congestion management "First, the local network must be approaching a congested state for our technique to even look for traffic to manage. Assuming that is the case, customers accounts must exceed a certain percentage of their upstream or downstream (both currently set at 70%) bandwidth for longer than a certain period of time, currently set at 15 minutes." |
|
bradyrYCCD - Network Operations Premium Member join:2008-10-27 Sonora, CA (Software) pfSense Netgear CM1200 Ubiquiti UAP-AC-HD
|
to inMaryland
from what i've read, comcast's congestion management practices don't do deep packet inspection (DPI). basically it doesnt care [or know] what protocols you're using/ what you are downloading.
now, of course that info could be out of date, and frankly it wouldn't surprise me if P2P traffic, for example, was a "priority target" over some other traffic, if you are on a congested node or otherwise...
also, it would be sweet to see some "evidence" of comcast's congestion management in practice...like show me some tcp graphs.. what does the data stream actually look like? ramp up then get knocked down, rinse and repeat (saw tooth) or is it ramp up, then get knocked down and stay down (capped) for the length of the download... |
|
jlivingood Premium Member join:2007-10-28 Philadelphia, PA 1 edit
1 recommendation |
to Crispyfertoo
That looks like normal TCP behavior...
If you want a flow showing shaping you want to see a table-top (flat) area reached over time, kind of like the pictures on the right hand side:
So... not throttling by us. Your traffic pattern looks fine.
|
|
jlivingood
1 recommendation |
to Crispyfertoo
Some more good charts for you to explain this pattern:
|
|
Xyc Premium Member join:2006-06-08 Sewell, NJ |
Xyc
Premium Member
2012-Oct-9 1:59 pm
So when do we get per-service-flow RED enabled to maximize bandwidth by throwing packets away? (» en.wikipedia.org/wiki/Ra ··· etection) |
|
|
to Crispyfertoo
That looks like speed boost. They give you a couple of minutes of increased speed to tease you then it goes down to the speed you paid for. |
|
EGThe wings of love Premium Member join:2006-11-18 Union, NJ
1 recommendation |
EG
Premium Member
2012-Oct-9 4:01 pm
That's not how it would look.. |
|
|
inMarland to bradyr
Anon
2012-Oct-9 4:09 pm
to bradyr
posting TCP graphs is a very good idea... I will do so. |
|
bradyrYCCD - Network Operations Premium Member join:2008-10-27 Sonora, CA (Software) pfSense Netgear CM1200 Ubiquiti UAP-AC-HD
1 edit |
to Crispyfertoo
entire stream | powerboost falloff |
okay, here are two tcp stream graphs of a 315MB file i copied from work to home (I have comcast blast...) feel free to ask any questions. one is the entire stream, and the other is zoomed in to show the powerboost fall-off. you can see that after the falloff, the remainder of the download is pretty much a straight line (on the graph) The fall off is after about 25 seconds, it drops down to 3.1-3.2MB/s sustained speed (approx 24Mbps) boosted is around 34Mbps.. |
|
|
to Crispyfertoo
Test with this linux torrent. It is a guaranteed ultra fast torrent. If it downloads normally it's not Comcast. » btguard.com/test.torrentDave |
|
|
inMaryland to bradyr
Anon
2012-Oct-9 9:01 pm
to bradyr
nntp download | http download |
nntp download graph |
|
inMaryland |
inMaryland
Anon
2012-Oct-10 9:19 am
Re: Is this throttling? - I believe soThe tcp graphs I posted yesterday indicate the download behavior I have been seeing for over a week now. Tcp traffic through NNTP (easynews and giganews) peaks at my bandwidth limit for a very short time then bounces around from zero to something aproaching my bandwidth limit (using multi threads)... When I allow the download to continue for a long period - the download averages about 18mbps - far below my limit of 105mbps.. This has been very consistent for over a week now.. and I have been using newsgroup downloads w/o issue for years.. I am absolutely NOT seeing this issue with other downloads... I have tested using large (non nntp) files and it consistently maxes out my bandwidth for the duration of the download. |
|
Extide join:2000-06-11 Salt Lake City, UT |
Extide
Member
2012-Oct-10 12:31 pm
Your nntp server simple appears to not have enough bandwidth available to serve you at 105mbit. There is likely some form of shaping going on at their end, not yours, in this case. |
|
|
inMaryland
Anon
2012-Oct-10 1:31 pm
I have been using easynews for years - they can easily fill the 105 mbps bandwidth - they have a very active support forum and there has been no complaint about throughput - I use 30 threads to download. In addition, as another test, I subscribed to a 14 day trial on giganews, using 30 threads from giganews results in the same sawtooth pattern. I am quite sure this is NOT an issue caused by the NNTP (usenet) source. To review, based on my testing during the past week, the slow download issue:
. is NOT caused by the NNTP source . happens only with NNTP downloads (I don't use p2p ie. torrents) . is NOT caused by my modem . is NOT caused by my router . is NOT related to node saturation . is NOTcaused by my individual over use of the bandwidth . I can recreate the results every time.. (slow (sawtooth) NNTP downloads, full throughput speed for http large file downloads) |
|
inMaryland |
inMaryland
Anon
2012-Oct-10 2:25 pm
easynews offers https ssl (port 81) downloads.. it is not optimal for my setup (can't use newsbin this way) but I will try this download method and report back... should be an interesting test. |
|
inMaryland |
inMaryland
Anon
2012-Oct-10 4:27 pm
I tested a usenet download from easynews using their https ssl port 81 method.. it works great.. maxed out my 105mbps bandwidth for the entire 1.4G download.. it is clear to me that Comcast is now traffic shaping... probably looking at packets for NNTP and maybe other types of traffic. |
|
JohnInSJ Premium Member join:2003-09-22 Aptos, CA |
JohnInSJ
Premium Member
2012-Oct-10 4:45 pm
said by inMaryland :I tested a usenet download from easynews using their https ssl port 81 method.. it works great.. maxed out my 105mbps bandwidth for the entire 1.4G download.. it is clear to me that Comcast is now traffic shaping... probably looking at packets for NNTP and maybe other types of traffic. Well, that's hardly an exhaustive test... |
|
|
inMaryland
Anon
2012-Oct-10 5:29 pm
did not say it was an exhaustive test.. I will keep trying both methods of downloading from easynews (NNTP vs. https) for awhile (days, weeks) and report back.. |
|
|
to Crispyfertoo
Re: Is this throttling?That looks completely normal if lots of upload activity and some rate limit/qos... As the download speed goes up the acks increase on the upstream until eventually the upstream becomes saturated (skyrocketing latency) and then throughput goes down the drain. Once the downstream goes down ACK's have gone down and then it starts to climb again repeating this process.
If you are no where near your upload capacity (like less than 50%) then I would say this isn't normal with something like bit torrent this can easily if your upload rate limit is not set high enough (not enough to include ACKs for your maximum download speed). |
|
jlivingood Premium Member join:2007-10-28 Philadelphia, PA |
to inMaryland
Re: Is this throttling? - I believe sosaid by inMaryland :did not say it was an exhaustive test.. I will keep trying both methods of downloading from easynews (NNTP vs. https) for awhile (days, weeks) and report back.. Fun experiment... What cable modem make/model and home gateway do you have? In your home gateway, any QoS settings for the upstream? If so, can you try limiting upstream bandwidth to 70% of your provisioned rate? (see » vonage.nmhoy.net/wrt54glqos.html but just try upstream limiting first before prioritizing FTP) Jason |
|