Geekzone: technology news, blogs, forums
Guest
Welcome Guest.
You haven't logged in yet. If you don't have an account you can register now.


View this topic in a long page with up to 500 replies per page Create new topic
1 | ... | 7 | 8 | 9 | 10 | 11 | 12 
mercutio
1387 posts

Uber Geek


  #584260 20-Feb-2012 23:13
Send private message

jpollock:
mercutio: 

have you tried a download accelerator?  axel is an example for linux.  not sure what is out for windows.



A wget replacement won't fix the problem for other TCP connections such as video streaming. :)

mercutio: 

i think you'll find that it's consistent low grade packet loss causing issues.



Yes, the impressive thing is that it doesn't affect UDP traffic, and limits sockets to the rate they would achieve at that ping without the large window extension.  That indicates that it is intentional behaviour of the network.  We've diagnosed it as a problem with Reach, Telstra Clear's upstream provider. :)

mercutio: 

the easiest solution is to bounce your traffic through a server in los angeles to your remote destination.  with tcp.  you can setup something like redir with a firewall only allowing connections to your ip... 

not sure if there is an easy way to do it transparently with vpn.  but i think you'll find vpn on it's own doesn't help.

 


Well, since LA is the wrong side of Reach, we still won't see any performance benefits, which is why you want a socket to a VPN provider here in NZ that uses a different international traffic provider, or else a VPN that uses UDP or load balanced TCP. :)  You could even implement some of Google's proposed TCP changes (quick start/etc) to reduce the number of round trips to the remote VPN endpoint. :)

I'd like my 10mbps per socket please! 

Jason 


interesting that it doesn't happen on udp.  i assumed tcp was reacting worse to the small amount of packet drops seen in udp.

you've done iperf testing with udp i take it?

what are you using to measure tcp/ip losses?  i'd be keen to try doing some testing of my own.

the thing about LA, is partially that it's closer.

wtf... i just tried to one of my LA hosts.. and it's going even worse than normal..

ok... my other LA host appears to be working fine through telstraclear... http://arp.meh.net.nz:24/10m

10mb test download.. peered with reach at equnix.

then another test that's slow

http://lv.meh.net.nz:24/10m 

peered with reach at any2ix.

the alternative port will bypass any port 80 transparent proxies 

 
 
 

GoodSync. Easily back up and sync your files with GoodSync. Simple and secure file backup and synchronisation software will ensure that your files are never lost (affiliate link).
Talkiet
4689 posts

Uber Geek

Trusted

  #584261 20-Feb-2012 23:13
Send private message

jnawk:
Talkiet: Mind you, you've taken some very interesting interpretations elsewhere in this thread so it's entirely possible you believe that a company using (for example) default settings out of the box means they deliberately and precisely understand and fully endorse the resulting behaviour, even if in fact the service is entirely fit for purpose and the cost of modifying it would far exceed any benefit it would bring them.

Cheers - N



Mate, lets not start a pissing match, but if you believe for one moment that telstra don't have capable people who understand their stuff, then you do them a disservice.


I never said that, I implied that even if there's a configuration somewhere which could be changed to improve things for you, that it probably doesn't merit the cost and time to do so, especially if it only affects a very very small number of people. And by affect, I mean actually is detrimental to their Internet use experience - not just that they may be having the same behaviour occur.

I know a number of past and current capable Telstra engineers and I would never accuse them of being unable to understand their stuff.

Like any Telco though, I would say it could prove very hard to get such a tiny change accepted, developed, tested, through change control and rolled out - especially when the revenue side of the change is likely to be Zero.

Cheers - N





--

 

Please note all comments are the product of my own brain and don't necessarily represent the position or opinions of my employer, previous employers, colleagues, friends or pets.


mercutio
1387 posts

Uber Geek


  #584264 20-Feb-2012 23:23
Send private message

http://mirrors.kernel.org/ubuntu-releases/releases/releases/11.10/ubuntu-11.10-server-amd64.iso

is a good example of something that's slow through telstraclear, fast through others, and on the close side of the US...

 



jnawk

176 posts

Master Geek


  #584267 20-Feb-2012 23:32
Send private message

Talkiet:Like any Telco though, I would say it could prove very hard to get such a tiny change accepted, developed, tested, through change control and rolled out -


Oh no problem there - Telstra don't even believe in testing their changes, much less CAB them! :-)

mercutio
1387 posts

Uber Geek


  #584278 21-Feb-2012 00:11
Send private message

it'd be nice to have an update from telstraclear.
 

jpollock
600 posts

Ultimate Geek

Trusted

  #584304 21-Feb-2012 07:37
Send private message

mercutio: http://mirrors.kernel.org/ubuntu-releases/releases/releases/11.10/ubuntu-11.10-server-amd64.iso

is a good example of something that's slow through telstraclear, fast through others, and on the close side of the US...

 


Careful, HTTP goes through the transparent proxy, which means that it can't be used to diagnose IP layer faults.

To really track it down, you'll need an encrypted connection and control of both ends to make sure that the large window size TCP extensions are being passed all the way through the network in both directions.

The testing here was done by comparing results between two NZ ISPs to a host in the US.  One ISP supported the large window option, another behaved like it didn't, but appeared to pass the data.




jnawk

176 posts

Master Geek


  #584312 21-Feb-2012 08:13
Send private message

jpollock:
mercutio: http://mirrors.kernel.org/ubuntu-releases/releases/releases/11.10/ubuntu-11.10-server-amd64.iso

is a good example of something that's slow through telstraclear, fast through others, and on the close side of the US...

?


Careful, HTTP goes through the transparent proxy, which means that it can't be used to diagnose IP layer faults.

To really track it down, you'll need an encrypted connection and control of both ends to make sure that the large window size TCP extensions are being passed all the way through the network in both directions.

The testing here was done by comparing results between two NZ ISPs to a host in the US. ?One ISP supported the large window option, another behaved like it didn't, but appeared to pass the data.


The same tests were also done to another host in Germany. Again, I had control of both ends. Same type of results - since the latency was even higher, the single socket speeds were even slower.



mercutio
1387 posts

Uber Geek


  #584374 21-Feb-2012 10:21
Send private message

jpollock:
mercutio: http://mirrors.kernel.org/ubuntu-releases/releases/releases/11.10/ubuntu-11.10-server-amd64.iso

is a good example of something that's slow through telstraclear, fast through others, and on the close side of the US...

 


Careful, HTTP goes through the transparent proxy, which means that it can't be used to diagnose IP layer faults.

To really track it down, you'll need an encrypted connection and control of both ends to make sure that the large window size TCP extensions are being passed all the way through the network in both directions.

The testing here was done by comparing results between two NZ ISPs to a host in the US.  One ISP supported the large window option, another behaved like it didn't, but appeared to pass the data.


tcpdump can show the window sizes pretty easily, if you can dump on both sides.  i think you'll find it's early packet loss stopping cubic from ramping up speed that really hurts.  then regular low amounts of packet loss stopping the speed going up too much.

 

jnawk

176 posts

Master Geek


  #584398 21-Feb-2012 10:55
Send private message

mercutio:

tcpdump can show the window sizes pretty easily, if you can dump on both sides. ?i think you'll find it's early packet loss stopping cubic from ramping up speed that really hurts. ?then regular low amounts of packet loss stopping the speed going up too much.

?


True, though the point is that with the transproxy in there, you can't be sure its not messing things up in its own special ways.

jpollock
600 posts

Ultimate Geek

Trusted

  #584405 21-Feb-2012 11:06
Send private message

jnawk:
mercutio:

tcpdump can show the window sizes pretty easily, if you can dump on both sides. ?i think you'll find it's early packet loss stopping cubic from ramping up speed that really hurts. ?then regular low amounts of packet loss stopping the speed going up too much.

?


True, though the point is that with the transproxy in there, you can't be sure its not messing things up in its own special ways.


Which it does by redoing the DNS lookup. ;) 




mercutio
1387 posts

Uber Geek


  #584415 21-Feb-2012 11:29
Send private message

jnawk:
mercutio:

tcpdump can show the window sizes pretty easily, if you can dump on both sides. ?i think you'll find it's early packet loss stopping cubic from ramping up speed that really hurts. ?then regular low amounts of packet loss stopping the speed going up too much.

?


True, though the point is that with the transproxy in there, you can't be sure its not messing things up in its own special ways.


well you can test that with alternative port.
 

mercutio
1387 posts

Uber Geek


  #584416 21-Feb-2012 11:30
Send private message

jpollock:
jnawk:
mercutio:

tcpdump can show the window sizes pretty easily, if you can dump on both sides. ?i think you'll find it's early packet loss stopping cubic from ramping up speed that really hurts. ?then regular low amounts of packet loss stopping the speed going up too much.

?


True, though the point is that with the transproxy in there, you can't be sure its not messing things up in its own special ways.


Which it does by redoing the DNS lookup. ;) 


i used to assume the reason telstraclear was show for web was it's transparent proxy.

these days i'm not convinced.

1 | ... | 7 | 8 | 9 | 10 | 11 | 12 
View this topic in a long page with up to 500 replies per page Create new topic





News and reviews »

New Air Traffic Management Platform and Resilient Buildings a Milestone for Airways
Posted 6-Dec-2023 05:00


Logitech G Launches New Flagship Console Wireless Gaming Headset Astro A50 X
Posted 5-Dec-2023 21:00


NordVPN Helps Users Protect Themselves From Vulnerable Apps
Posted 5-Dec-2023 14:27


First-of-its-Kind Flight Trials Integrate Uncrewed Aircraft Into Controlled Airspace
Posted 5-Dec-2023 13:59


Prodigi Technology Services Announces Strategic Acquisition of Conex
Posted 4-Dec-2023 09:33


Samsung Announces Galaxy AI
Posted 28-Nov-2023 14:48


Epson Launches EH-LS650 Ultra Short Throw Smart Streaming Laser Projector
Posted 28-Nov-2023 14:38


Fitbit Charge 6 Review 
Posted 27-Nov-2023 16:21


Cisco Launches New Research Highlighting Gap in Preparedness for AI
Posted 23-Nov-2023 15:50


Seagate Takes Block Storage System to New Heights Reaching 2.5 PB
Posted 23-Nov-2023 15:45


Seagate Nytro 4350 NVMe SSD Delivers Consistent Application Performance and High QoS to Data Centers
Posted 23-Nov-2023 15:38


Amazon Fire TV Stick 4k Max (2nd Generation) Review
Posted 14-Nov-2023 16:17


Over half of New Zealand adults surveyed concerned about AI shopping scams
Posted 3-Nov-2023 10:42


Super Mario Bros. Wonder Launches on Nintendo Switch
Posted 24-Oct-2023 10:56


Google Releases Nest WiFi Pro in New Zealand
Posted 24-Oct-2023 10:18









Geekzone Live »

Try automatic live updates from Geekzone directly in your browser, without refreshing the page, with Geekzone Live now.



Are you subscribed to our RSS feed? You can download the latest headlines and summaries from our stories directly to your computer or smartphone by using a feed reader.







Lenovo