dslreports logo
site
    All Forums Hot Topics Gallery
spc
Search Topic:
uniqs
4696
share rss forum feed

resa1983
Premium
join:2008-03-10
North York, ON
kudos:10
reply to TSI Gabe

Re: Youtube

We randomly have the buffer issue with youtube, where it refuses to buffer at all.

Right now, I'm running the pytomo program, while my son is watching a youtube video which isn't having any problems buffering (probably because he watches the same video several times a day?).
--
Battle.net Tech Support MVP


jmck
formerly 'shaded'

join:2010-10-02
Ottawa, ON

does he watch them in HD too? when i had issues with TSI/youtube reloading the page multiple times would eventually make it go through.

i also had tried using youtube downloaders and the actual HTTP download would time out too.


stevey_frac

join:2009-12-09
Cambridge, ON
reply to TSI Gabe

Where are we supposed to upload the data to?

Email Gabe? or?

EDIT: Nevermind, support@teksavvy.com



TSI Martin
Premium
join:2006-02-23
Chatham, ON
kudos:33

said by stevey_frac:

Where are we supposed to upload the data to?

EDIT: Nevermind, support@teksavvy.com

You can also upload it to the »TekSavvy Direct. We'll forward it.

Martin
--
TSI Martin (Escalations / E-Services) - TekSavvy Solutions Inc.
Authorized TSI employee ( »»TekSavvy FAQ »Official support in the forum )
Follow us on Twitter : @TekSavvyCSR ; @TekSavvyNetwork

wenk

join:2008-09-02
Ottawa, ON

OK, this is an interesting result. I'm using a download plugin for firefox to download a video. Something less popular, will only get a few thousand views. It's been downloading at 150-400KByte/s. Mostly 150, sometimes climbing up to 400+.

I load husky's latest video (watch?v=69YAadYgMj4&feature=youtube_gdata) Not only does it load at close to 1MByte/s the video that I'm downloading also speeds up to >1Mbyte/s. I was able to repeat this a couple of times. This would explain why videos will download quickly at say 6PM when I get home, but if I have to reload the video later (ie when Chrome decides to wipe the video cache) it will load very slowly. Typically when I get home I'm pre-buffering several videos many of them from more popular sources (ie The Game Station sources, Yogscast etc).


graniterock

join:2003-03-14
London, ON
Reviews:
·WIND Mobile
·TekSavvy Cable
reply to TSI Martin

said by TSI Martin:

You can also upload it to the »TekSavvy Direct. We'll forward it.

Martin

I ended up doing this. I never got a confirmation from the email that it got through ok. The attachment was 5.5 MB so I wonder if it go rejected?


TSI Martin
Premium
join:2006-02-23
Chatham, ON
kudos:33

PM the email you sent it from... I'll confirm if we did receive it or not.



creed3020
Premium
join:2006-04-26
Kitchener, ON
kudos:2
reply to TSI Gabe

I sent my results in. It was a rather large attachment as well. Confirmation of receipt would be great.


Scycotic

join:2012-12-10

Hm, mine was only 99 KB, but then I only ran it for 5 hours or so.


graniterock

join:2003-03-14
London, ON

My 48+ hour run generated a 1.2 MB .db file and a 4 MB compressed log file (50 MB uncompressed!).

Expand your moderator at work

wenk

join:2008-09-02
Ottawa, ON
reply to graniterock

Re: Youtube

So is there any update on this?

Again hitting different ggc servers will actually increase the speed of other ggc servers. I've been using this successfully when youtube decides to slow to a crawl, but it burns through a lot of bandwidth.


stevey_frac

join:2009-12-09
Cambridge, ON

I was using YouTube as a playlist for music for my son. Videos kept pausing and halting. Kinda hard to explain to a two year old that, ya it IS broken, but no, daddy can't fix it.

Any update?


Majromax

join:2012-09-02
Dollard-Des-Ormeaux, QC
reply to TSI Gabe

I did a 24h pytomo analysis from my home system over the past day. The database (.db and .tbz files) have been e-mailed to support@teksavvy.com. Using the web interface from pytomo, I was able to find out a few interesting things that might be of broader interest:

First, pytomo by default only appears to download standard-definition videos. The video encoding rate was uniformly in the 600kbit/sec range, which corresponds to about 480p videos (youtube's 'show video info' widget, by the way, doesn't give good encoding readings for 720 and 1080p). Consequently and unsurprisingly, on my lightly-loaded 25/10 connection I never saw buffering.

There was an interesting statistic, however, in the InitialRate data -- that represents the burst-download initial buffering of the first few seconds of the video. Overall, the burst rate was 12.5mbit/sec for me (roughly half of my line rate), which seems fine, but that was heavily modified by congestion. The burst rate fell from 12.5mbit/sec to about 6 from 18h-22h (eastern time), and then it recovered back to its full speed between 22h and 24h as peak congestion let up.

Checking the downloaded videos, these were all accessed from Toronto-based cache servers (all but one video seemed to be on Teksavvy's internal Youtube cache servers / googleboxes, at that.)

The peak-hours congestion can easily be causing the buffering that people here see. Based on youtube's upload quality recommendations, "standard quality" 1080p video should be encoded at 8mbit/sec. At this original bandwidth, the congested-peak-rate won't be enough for unbuffered playback. Youtube does appear to re-encode these videos to about 3.5mbit/sec, but I don't know the requirements for doing so -- this should be playable at congested hours, but there isn't much headroom.

In terms of other network diagnostics, nothing pointed to congested lines themselves. The average ping varied between 10ms and about 15, and it was only weakly related to the congested period.

Unfortunately, I could only test over IPv4. I do have an IPv6 login, but I found out only after signing up for the IPv6 beta that DD-WRT doesn't properly support IPv6-over-pppoe.

TL;DR: Teksavvy-network Youtube cache servers in Toronto are overloaded between about 8pm and 11pm EST, to the point that 1080p videos may require buffering. It's a datapoint of one, so I have to be right, eh?


singsonn

join:2009-08-12
Montreal, QC
reply to TSI Gabe

I have IPv6 configured on my Linux Debian box. How do I select to perform the test in IPv4 only? The tests seems to run in IPv6 at the moment :

root@debian:/opt/pitomo/Pytomo-2.2.4# ps auxwww
[...]
root 32660 0.3 8.5 61548 22380 pts/0 S+ 11:21 0:01 python ./start_crawl.py
[...]
root@debian:/opt/pitomo/Pytomo-2.2.4# netstat -tanup | grep 32660
tcp6 0 0 2607:f2c0:f00f:f8:42733 2607:f8b0:400d:c00:::80 ESTABLISHED 32660/python

I think it would give more accurate results.

I know I could simply disable IPv6, but I would like to know if there is an option to just force IPv4 with this test.


Majromax

join:2012-09-02
Dollard-Des-Ormeaux, QC

An IPv6 test would probably itself be interesting, it might show a different congestion pattern than IPv4.

Looking through the pytomo source, it looks like the easiest way to disable IPv6 would be to tweak the DNS library to pretend it can't use INET6 sockets. I don't have any testing available, but editing pytomo/dns/inet.py to read:

#try: # Pretend that we don't have INET6 at all
# AF_INET6 = socket.AF_INET6
#except AttributeError:
AF_INET6 = 9999


beginning at line 34 might do the trick. The resolving will still happen, but it should pretend that IPv6 addresses can't be processed (failing back to ipv4).

That, or it may break things utterly. Either/or.



enzymes

join:2003-11-29
Brampton, ON
Reviews:
·TekSavvy Cable
reply to TSI Gabe

So I came across this guide for better youtube performance...

»mitchribar.com/2013/02/how-to-st···s-guide/

will it work for teksavvy/rogers as well?