Problems at end of slow satelite link
Newbe hoping this is the correct area to post.
I've been reading the LARTC HOWTO, but at 110 pages, I am finding it difficult to know where to start. I hope you may be able to surgest a starting place to a couple of problems I am having.
I am dealing with the slow end of a satelite ADSL. This is 'lumpy', with huge latency (1 to 30 sec), 10% dropped packets, huge but
intermittent bursts of packets, and generally a pile of poo. The
supplier (BT) is of course no help, so we have to live with what we have.
I have tried limiting the bandwidth with a Tocket Bucket Filter with almost no effect.
What I think I need is to agressivelly schedule SSH and Telnet before all other TCP connections. Others I don't care about. There are I see many scheduling options, can any member surgest how I might atchive this?
The second problem I have is that my connection keeps terminating (Telnet, SSH) with 'Connection closed by foreign host', with no message in logs. Always during streaming of data. Some part of the kernel is getting tired of waiting for the other end to respond, or possible a buffer overflow? Again, if any member may surgest how this can be countered with extream predudice, I would be extreamly greatful.
Regards, Ben Clewett.
_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
Quoted 18 lines · Ben Clewett
Newbe hoping this is the correct area to post.
I've been reading the LARTC HOWTO, but at 110 pages, I am finding it difficult to know where to start. I hope you may be able to surgest a starting place to a couple of problems I am having.
I am dealing with the slow end of a satelite ADSL. This is 'lumpy', with huge latency (1 to 30 sec), 10% dropped packets, huge but
intermittent bursts of packets, and generally a pile of poo. The
supplier (BT) is of course no help, so we have to live with what we have.I have tried limiting the bandwidth with a Tocket Bucket Filter with almost no effect.
What I think I need is to agressivelly schedule SSH and Telnet before all other TCP connections. Others I don't care about. There are I see many scheduling options, can any member surgest how I might atchive this?
Quoted 6 lines
The second problem I have is that my connection keeps terminating (Telnet, SSH) with 'Connection closed by foreign host', with no message in logs. Always during streaming of data. Some part of the kernel is getting tired of waiting for the other end to respond, or possible a buffer overflow? Again, if any member may surgest how this can be countered with extream predudice, I would be extreamly greatful.
I think it's best you create a htb setup with different classes. One of the classes needs a higher priority so the packets in that classes are send first. Next you need to put the ssh/telnet/syn/ack packets in that classes. All needed information can be found in the LARTC howto. And you can also find more info on www.docum.org.
Stef
Signature · 8 lines
[email protected] "Using Linux as bandwidth manager" http://www.docum.org/ #lartc @ irc.oftc.net _______________________________________________ LARTC mailing list / [email protected] http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
Quoted 5 lines · Stef Coene
I think it's best you create a htb setup with different classes. One of the classes needs a higher priority so the packets in that classes are send first. Next you need to put the ssh/telnet/syn/ack packets in that classes. All needed information can be found in the LARTC howto. And you can also find more info on www.docum.org.
This is Hierarchical Token Bucket, explained in section 9.4.5 of the HOWTO document?
I was rather hoping you were going to surgest something else, as this is not in my kernel... :)
I will however check out the documentation in some more detail. There are a few unresolved questions I have on this pleasent protocol.
Like I notice it claims to scale up to the available bandwidth in proporsion to the allocated bandwidth. I need to know how it calculates this bandwidth, as our link shows all the properties of the available bandwidth being proporsional to 1/S noise. Ie, unpredictable and un-averagable.
I would like to know also whether HTB scales down, if bandwidth becomes throtted by our unpredictable pipe. And in either case, whether it's possible to have a protected channel, like an admin channel, which doen't scale up or down with the rest, staying at, say, 16kbit...
All of which I will now try and find out. And then attempt to get it into my kernels at either end. :)
Ben
The biggest mistory of my link-from-hell is the ping. This shows either a latency of 800ms, or 1.5 seconds, or up to 30 seconds when the line goes on-hold for a while. About once every two minutes. Yet Telnet/SSH consistently give latency far far more than this, of about 10 to 60 seconds consistetly.
If anybody can surgest a reason for this, I think this is likelly to be at the heart of any fix I can find...
Regards again...
Quoted 3 lines
Stef
_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
Quoted 7 lines · Ben Clewett
Quoted 5 lines · Stef Coene
I think it's best you create a htb setup with different classes. One of the classes needs a higher priority so the packets in that classes are send first. Next you need to put the ssh/telnet/syn/ack packets in that classes. All needed information can be found in the LARTC howto. And you can also find more info on www.docum.org.
Quoted 5 lines
This is Hierarchical Token Bucket, explained in section 9.4.5 of the HOWTO document?
I was rather hoping you were going to surgest something else, as this is not in my kernel... :)
You can also use cbq. But htb is easier to configure. Or go for the 2.4.20 kernel. That has htb support.
Quoted 9 lines
I will however check out the documentation in some more detail. There are a few unresolved questions I have on this pleasent protocol.
Like I notice it claims to scale up to the available bandwidth in proporsion to the allocated bandwidth. I need to know how it calculates this bandwidth, as our link shows all the properties of the available bandwidth being proporsional to 1/S noise. Ie, unpredictable and un-averagable.
Quoted 4 lines
I would like to know also whether HTB scales down, if bandwidth becomes throtted by our unpredictable pipe. And in either case, whether it's possible to have a protected channel, like an admin channel, which doen't scale up or down with the rest, staying at, say, 16kbit...
You can configure a protected channel with a minimum bandwidth.
Quoted 10 lines
All of which I will now try and find out. And then attempt to get it into my kernels at either end. :)
Ben
The biggest mistory of my link-from-hell is the ping. This shows either a latency of 800ms, or 1.5 seconds, or up to 30 seconds when the line goes on-hold for a while. About once every two minutes. Yet Telnet/SSH consistently give latency far far more than this, of about 10 to 60 seconds consistetly.
Have you tried screen? If you start screen, you get a virtual console. You can work in it like you normally do, but if your session is terminated, you can reconnect to the same screen session with "screen -r". Very handy if you have a faulty connection :)
Quoted 2 lines
If anybody can surgest a reason for this, I think this is likelly to be at the heart of any fix I can find...
I have no idea ..
Stef
Signature · 8 lines
[email protected] "Using Linux as bandwidth manager" http://www.docum.org/ #lartc @ irc.oftc.net _______________________________________________ LARTC mailing list / [email protected] http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/