MTU Changed

[minkukel]

Anybody noticed the change in MTU setting ?
Installed IPCOP this morning. Got fed up wiff SWE.
And when doing a ifconfig ppp0 I noticed that the MTU is set at 1432 after the pppd moaned about cant set MTU to 1500. Average combined download speed is up from 12 kb/s to 50kb/s.

 

[Gatecrasher]

IC and I discussed this here:

http://www.mybroadband.co.za/vb/showthread.php?t=16486&page=2

I've changed mine to 1432, but things are so bad right now, I don't think it makes much difference...

I can still get a decent multithreaded download (50-60KB/s - I guess its like using a blunderbus - you can't really miss) but everthing else is terribly painful and sluggish. Getting 100-200kbps on the Telkom speed test. And disconnects every couple of hours. Up until Friday night's authentication problem, everything was singing along at full speed.

So something has happened, and WBS are either not letting on or they haven't a clue.

My pings to Iburst are excellent (all around 60ms with no errors), so I'm not having any signal problems. My pings to everything beyond Iburst (even local) are full of errors. That seems to confirm that there are serious problems between WBS and UUNET, since UUNET doesn't appear to be having any problems.

I don't believe they are throttling the service to save money. Way too early to be that stupid. If they had wanted to cut down on bandwidth they would have introduced the caps already.

 

[ic]

There's no need for a bandwidth cap when there is no speed.

 

[aborg]

Gatecrasher - I changed my MTU and RWIN settings to yours and things seem to be better, averaging about 60 - 80 KB/s on multithreads where before it was barely 20 KB/s. Don't know how long this will last though.

 

[ic]

MTU should be the same for all iBursters , although each base-station could in theory have IP routers with different MTU values, if that's true then everyone connecting via the same base-station should use the same MTU.

RWIN is more complicated, to find your optimal RWIN value you need to determine your connection's latency, that could in theory vary from one connection to the next even on the same base-station, even if it doesn't, the bursting nature of iBurst or how WBS have implemented it, seems to give wildly varying latency. Until the iBurst network stabilises ( ) we won't really be able to do more than guess RWIN values.

 

[Gatecrasher]

Until the iBurst network stabilises ( ) we won't really be able to do more than guess RWIN values.

Yeah, latency is all over the place.

The only good piece of tech I could find on this suggests using an RWin which is an even multiple of the MSS.

So RWIN = 2* (1 + INT[0.5*L*S/M])*M

The values I used are:
L (latency in ms) = 180ms
S (maximum connection speed) = 128KB/s
M (MSS) = 1404

which gives an RWIN of 25272, (18 x 1404)

Some texts suggest using much larger values. The drawback of that is if you are getting errors, downloads could slow up dramatically. That's because larger data packets will have to be resent. On the other hand, too small a value when everything is sweet, and you could get a lot of dead time waiting for an acks between packets.