Packet Loss (BitTorrent Protocol)

ginggs said:
How exactly are you doing that? Why is there so much collateral damage?

Why are other completely unrelated services like Skype, SSH and Steam affected?

Why are torrents from legitimate services like the Blizzard Updater (World of Warcraft and Starcraft II) affected?
Why doesn't iBurst set up a bittorrent client permanently connected to the Blizzard trackers for peering with iBurst subscribers? This should save you a bit in international traffic and improve the experience for your subscribers.
Click to expand...

We use industry leading service control engines. The issue is not bandwidth but resources (time slots) on the base station. A peer to peer client with an unlimited number of connections hogs the base station, not because of bandwidth but connection count. If it was not shaped then normal (non-torrenting) users would have their performance badly affected. We would rather have a network that provides banking, informational and email services than a clogged downloading service. Using HTTP and and other alternatives like NTP are a workable alternative.

I am unaware of Skype or SSH being affected as we use those protocols extensively and don't have an issue. I have a daily Skype conference to Cape Town on iBurst using UTIDs and haven't experienced any problems.
 
eltherza said:
You can download WoW updates via normal http, which in most cases is always faster and less bandwidth intensive [no uploading]. It's a better solution. I don't know about starcraft 2 though.

My observations is that iBurst has a limit on the number of active connections at once. Torrents open alot of connections. if you can, try limit the number of concurrent connections or else you may find it will drop some to make space [resulting in packet loss]. This is very observable when you hit the 64 VAS, to a point where you can not have 2 PCs browsing the internet at the same time (if you monitor it, only 1 will ever be loading a web page at a given time.)
Click to expand...

Good analysis.
 
eltherza said:
You can download WoW updates via normal http, which in most cases is always faster and less bandwidth intensive [no uploading]. It's a better solution. I don't know about starcraft 2 though.
Click to expand...
If you scratch around you can find the Starcraft II patches on HTTP as well, although none of these are official sources and therefore carry the associated risks. It would be nice if iBurst hosted their own mirror for these updates.
eltherza said:
My observations is that iBurst has a limit on the number of active connections at once. Torrents open alot of connections. if you can, try limit the number of concurrent connections or else you may find it will drop some to make space [resulting in packet loss]. This is very observable when you hit the 64 VAS, to a point where you can not have 2 PCs browsing the internet at the same time (if you monitor it, only 1 will ever be loading a web page at a given time.)
Click to expand...
Very interesting. It seems to me that some HTTP trackers are blocked and just about all DHT traffic is blocked as well, although this could just be a side-effect of the limit on the number of connections.
r00igev@@r said:
We use industry leading service control engines. The issue is not bandwidth but resources (time slots) on the base station. A peer to peer client with an unlimited number of connections hogs the base station, not because of bandwidth but connection count.
Click to expand...
How many connections are we allowed?
r00igev@@r said:
I am unaware of Skype or SSH being affected as we use those protocols extensively and don't have an issue.
Click to expand...
Read further back in this thread, the issue is while a BitTorrent client is open other services are affected. You cleared this up with your explanation about the connection count though.
 
ginggs said:
If you scratch around you can find the Starcraft II patches on HTTP as well, although none of these are official sources and therefore carry the associated risks. It would be nice if iBurst hosted their own mirror for these updates.

Very interesting. It seems to me that some HTTP trackers are blocked and just about all DHT traffic is blocked as well, although this could just be a side-effect of the limit on the number of connections.

How many connections are we allowed?

Read further back in this thread, the issue is while a BitTorrent client is open other services are affected. You cleared this up with your explanation about the connection count though.
Click to expand...

4 with identified peer to peer otherwise limit is not set (e.g. you can use 10 threads on DownThemAll!)
 
Why is P2P so slow on iBurst?

So I have been trying to download songs etc using limewire (this is not the issue, so dont go OT) and no matter what I download, even the most popular of songs, I never seem to top 8KB/s.

This happens at all time of the day, weekdays and weekends. But if I use even a 384DSL connection, the same song tops out at 41KB/s. Two things I have noticed though:

1. With iBurst it takes about 2-3 minutes to connect to the three required hosts, as it just keeps connecting and disconnecting instantly and only 2-3 minutes later it finds three stable hosts. On DSL this process take about 30-45 seconds.

2. The amount of dropped packets I get on iBurst is crazy. If you look at "Dropped (I/O)" once I get those 3 stable connections the dropped I/O averages about (70/0) even though I have 100% iburst signal strength. On DSL I havent checked, but Im sure this ratio is lower.

3. On DSL when downloading a song it says "Downloading at 41KB/s from 8 p2p users" while on iBurst it says "Downloading at 4KB/s" from 2 p2p users".

Why is this? It cant be shaping since I have tried this in the evenings and weekends as well. So what the hell is the problem???
 
Just_Ice said:
So I have been trying to download songs etc using limewire (this is not the issue, so dont go OT) and no matter what I download, even the most popular of songs, I never seem to top 8KB/s.
Click to expand...
Yes, it is.

From http://www.limewire.com/about:
LimeWire uses the BitTorrent protocol and the Gnutella network to provide unparalleled searches and download speed to the user.
Click to expand...
 
I have a N95 that took a trip off my roof when I was waterproofing and cracked. Interested?

Any takers for suggesting to relax P2P a bit between 2am and 6am? It's not much but looks feasible without stopping my granny from reading her emails.
 
r00igev@@r said:
I have a N95 that took a trip off my roof when I was waterproofing and cracked. Interested?
Click to expand...
Yes please!
r00igev@@r said:
Any takers for suggesting to relax P2P a bit between 2am and 6am? It's not much but looks feasible without stopping my granny from reading her emails.
Click to expand...
You mean relaxing the P2P shaping between 2am and 6am? Mental-Tree suggested something along those lines recently here.
 
As long as its after 1am I am happy, some people are still doing online gaming/raiding up to 1am (12 in europe)
 
Last edited:
ginggs said:
Yes, it is.

From http://www.limewire.com/about:
Click to expand...
I was unaware that P2P and the whole bit torrent network is shaped 24/7... It explains the shocking speeds though.

r00igev@@r said:
Any takers for suggesting to relax P2P a bit between 2am and 6am? It's not much but looks feasible without stopping my granny from reading her emails.
Click to expand...
Me!! Turning off or relaxing the shaping on P2P and bit torrents for at least a couple hours a day, even if it is between 12 and 6 would be much appreciated by many. And its not like there would be a massive spike in traffic because of it, as many of us still need to look after our caps.
 
Just_Ice said:
Me!! Turning off or relaxing the shaping on P2P and bit torrents for at least a couple hours a day, even if it is between 12 and 6 would be much appreciated by many. And its not like there would be a massive spike in traffic because of it, as many of us still need to look after our caps.
Click to expand...

The reason for the shaping has little to do with throughput but connection count. It results in unfair resource usage in a similar fashion to what a Smartphone does do a cellular network. The tuning would be to find a level that prevents this from occurring.
 
r00igev@@r said:
The reason for the shaping has little to do with throughput but connection count. It results in unfair resource usage in a similar fashion to what a Smartphone does do a cellular network. The tuning would be to find a level that prevents this from occurring.
Click to expand...
Oh I see... that would then explain why I can get 8 connections easily with DSL but with Iburst you are lucky to get 2.
Is there a fixed limit to the number of connections per tower?
 
r00igev@@r said:
All radio technologies have very similar usage restraints. The limit is dependant on a number of physical and logical parameters. The answer would be at 25 the service would be excellent and at 500 it would be very poor.
Click to expand...

Then, why do we get 2 and horrible service?
 
You must log in or register to reply here.
Back
Top
Sign up to the MyBroadband newsletter
X