Classful Queuing
OK, I'm stumped. I've read through most of the LARTC HOWTO and have yet to find a basis for what I need to accomplish.
I have a Linux box that controls access to and from the Internet at my workplace. We have a number of remote employees that connect via PPTP and IPSEC to the office's internal network. Some of these remote employees are currently using SIP phones. The problem is occasionally the available bandwidth becomes consumed and the VoIP calls (obviously) suffer because of this.
Configuration:
eth0 - connected to the internal office
eth1 - connected to the internet
pppX - incoming on eth1 connection
ipsec0 - incoming on eth1 connection
My question, how can I set classful htb queuing up so that it's rules encompass all traffic on eth1 (including that to and from the ipsec and ppp connections) while reserving bandwidth for and prioritizing the SIP traffic?
In looking through chapter 9 it appears that all the configurations apply to a specific interface, and thus would only get eth1 for example. While the traffic on the ppp and ipsec connections would arrive on the eth1 interface only after being placed on their specific interfaces and encrypted, thus most likely missing proper classification and prioritization.
Am I over thinking this problem or missing something?
I'll happily provide any clarification or additional information needed.
Signature · 8 lines
Jamin W. Collins It has always been Debian's philosophy in the past to stick to what makes sense, regardless of what crack the rest of the universe is smoking. -- Andrew Suffield _______________________________________________ LARTC mailing list / [email protected] http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
Hi
I think what you need to look for is marking of packets with netfilter - let it classify and then used tc to place the properly marked packets into the proper queue
Because you can mark in the PREROUTING table in mangle before it is enc
Alex
Quoted 42 lines · Jamin W. Collins
OK, I'm stumped. I've read through most of the LARTC HOWTO and have yet to find a basis for what I need to accomplish.
I have a Linux box that controls access to and from the Internet at my workplace. We have a number of remote employees that connect via PPTP and IPSEC to the office's internal network. Some of these remote employees are currently using SIP phones. The problem is occasionally the available bandwidth becomes consumed and the VoIP calls (obviously) suffer because of this.
Configuration:
eth0 - connected to the internal office
eth1 - connected to the internet
pppX - incoming on eth1 connection
ipsec0 - incoming on eth1 connectionMy question, how can I set classful htb queuing up so that it's rules encompass all traffic on eth1 (including that to and from the ipsec and ppp connections) while reserving bandwidth for and prioritizing the SIP traffic?
In looking through chapter 9 it appears that all the configurations apply to a specific interface, and thus would only get eth1 for example. While the traffic on the ppp and ipsec connections would arrive on the eth1 interface only after being placed on their specific interfaces and encrypted, thus most likely missing proper classification and prioritization.
Am I over thinking this problem or missing something?
I'll happily provide any clarification or additional information needed.
Signature · 8 lines
Jamin W. Collins It has always been Debian's philosophy in the past to stick to what makes sense, regardless of what crack the rest of the universe is smoking. -- Andrew Suffield _______________________________________________ LARTC mailing list / [email protected] http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
Quoted 6 lines · Alexander Samad
I think what you need to look for is marking of packets with netfilter - let it classify and then used tc to place the properly marked packets into the proper queue
Because you can mark in the PREROUTING table in mangle before it is enc
But will the mark still exist after the encryption/encapsulation? Plus, that only deals without outgoing not incoming utilization if I'm following this correctly. The incoming packets would not be identifiable until they were taken off the eth1 interface for their specific interface (ppp or ipsec), right?
Signature · 7 lines
Jamin W. Collins "Never underestimate the power of very stupid people in large groups." -- John Kenneth Galbraith _______________________________________________ LARTC mailing list / [email protected] http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
Quoted 6 lines · Jamin W. Collins
Quoted 5 lines · Alexander Samad
I think what you need to look for is marking of packets with netfilter - let it classify and then used tc to place the properly marked packets into the proper queue
Quoted 7 lines
Quoted 1 line
Because you can mark in the PREROUTING table in mangle before it is enc
But will the mark still exist after the encryption/encapsulation? Plus, that only deals without outgoing not incoming utilization if I'm following this correctly. The incoming packets would not be identifiable until they were taken off the eth1 interface for their specific interface (ppp or ipsec), right?
not so about ingres, but the marking stay with the packet after the enc ( well on 2.6 with native stack it does). I use this for marking packets.
Quoted 10 lines
Signature · 8 lines
-- Jamin W. Collins "Never underestimate the power of very stupid people in large groups." -- John Kenneth Galbraith _______________________________________________ LARTC mailing list / [email protected] http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
This is a multipart message in MIME format.
--=_alternative 0014B16386256F2B_Content-Type: text/plain; charset="US-ASCII"
Quoted 4 lines
But will the mark still exist after the encryption/encapsulation?
Quoted 3 lines
not so about ingres, but the marking stay with the packet after the enc ( well on 2.6 with native stack it does). I use this for marking packets.
Isn't this going to depend on whether you are encrypting the whole packet (VPN style) or just the data portion of the packet (SSL style)?
--=_alternative 0014B16386256F2B_Content-Type: text/html; charset="US-ASCII"
<br><font size=2><tt>>But will the mark still exist after the encryption/encapsulation?</tt></font> <br><font size=2><tt>>>not so about ingres, but the marking stay
with the packet after the enc<br>
>>( well on 2.6 with native stack it does). I use this for
marking<br>
>>packets.</tt></font>
<br>
<br><font size=2 face="sans-serif">Isn't this going to depend on whether
you are encrypting the whole packet (VPN style) or just the data portion
of the packet (SSL style)?</font>
<br>
--=_alternative 0014B16386256F2B_=--
_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
Quoted 16 lines · Jamin W. Collins
OK, I'm stumped. I've read through most of the LARTC HOWTO and have yet to find a basis for what I need to accomplish.
I have a Linux box that controls access to and from the Internet at my workplace. We have a number of remote employees that connect via PPTP and IPSEC to the office's internal network. Some of these remote employees are currently using SIP phones. The problem is occasionally the available bandwidth becomes consumed and the VoIP calls (obviously) suffer because of this.
Configuration:
eth0 - connected to the internal office
eth1 - connected to the internet
pppX - incoming on eth1 connection
ipsec0 - incoming on eth1 connection
Quoted 4 lines
My question, how can I set classful htb queuing up so that it's rules encompass all traffic on eth1 (including that to and from the ipsec and ppp connections) while reserving bandwidth for and prioritizing the SIP traffic?
First of all: policing ("shaping" incomming) does not really work as desired. Not even with RED. The dropping of packets causes to much retransmits, which will fill up your incomming twice. One could use RED/ECN, but till now i did not get RED to mark any packets with ECN. However.
Second: look at /etc/protocols or at tcpdump to identify the protocols you want to priorize and shape (not police). Use iptables, MARK and the -p option for that. Besides, I don't know whether you have more than one static IP. If you have more, you could set up "aliases" and shape via outgoing/source (and incomming/destination if you really want to police).
Signature · 10 lines
Robert Felber (EDV-Leitung) Autohaus Erich Kuttendreier Drosselweg 21 81827 Muenchen Tel: +49 (0) 89 / 453 12-86 Fax: +49 (0) 89 / 453 12-80 PGP: 896CF30B PGP-Fingerprint: CF36 AA93 9716 63E8 962F 15CC A80E 1A79 BF77 25EA
Quoted 7 lines · On Mon, Oct 11, 2004 at 10:46:01PM -0500, [email protected] wrote:
Quoted 4 lines
But will the mark still exist after the encryption/encapsulation?
Quoted 3 lines
not so about ingres, but the marking stay with the packet after the enc ( well on 2.6 with native stack it does). I use this for marking packets.
Isn't this going to depend on whether you are encrypting the whole packet (VPN style) or just the data portion of the packet (SSL style)?
I use it to mark parkets that are then esp enc. I am using in currently with 2.6 and native ipsec stack to mark all packets that come in as esp and then are de - enc, I allow these through the firewall. This was my way around the old the problem of how to setup the firewall when the ipsecX interface dissappeared.
I beleive the packet is encaped in place not duplicate. Then the new packet is refeed back in to netfilter.
Alex