Re: [c-nsp] Needs some help with QOS
I have crafted and applied some rules which I thought would prioritize traffic from an 871w (via ADSL) to one specific host. The idea is that any traffic destined to this host should be prioritized over all other traffic. What is your upstream connection? If you're using PPPoE, you won't be able to do any output queuing, as the outbound LAN interface is never saturated (the bottleneck is experienced by the DSL modem). Ivan ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
Re: [c-nsp] Needs some help with QOS
Exactly true ... That would be my next answer :) However, the problem is that it's somewhat hard to estimate what the shaping bandwidth should be in DSL environments (you have the cell tax on top of PPPoE plus unknown amount of oversubscription in the SP network) if you want to squeeze as much out of the DSL line as possible. Best regards Ivan -Original Message- From: Tim Franklin [mailto:t...@pelican.org] Sent: Tuesday, March 24, 2009 1:57 PM To: Ivan Pepelnjak Cc: 'John Lange'; 'Cisco NSP' Subject: Re: [c-nsp] Needs some help with QOS On Tue, March 24, 2009 12:12 pm, Ivan Pepelnjak wrote: What is your upstream connection? If you're using PPPoE, you won't be able to do any output queuing, as the outbound LAN interface is never saturated (the bottleneck is experienced by the DSL modem). If you know what your upstream bandwidth is, you can wrap a shaper around the queueing policy to provide the back-pressure. Useful for all sorts of 'ethernet hand-off' type services where the circuit provider has some other device upstream of your router. Regards, Tim. ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
Re: [c-nsp] Needs some help with QOS
Hi, you should use hierarchical QoS. First of all you should shape the output traffic down to the upstream speed, then you can use the llq inside the shaped class: http://www.cisco.com/en/US/tech/tk543/tk545/technologies_tech_note09186a00800b2d29.shtml BR, A. On Tue, 24 Mar 2009, Ivan Pepelnjak wrote: I have crafted and applied some rules which I thought would prioritize traffic from an 871w (via ADSL) to one specific host. The idea is that any traffic destined to this host should be prioritized over all other traffic. What is your upstream connection? If you're using PPPoE, you won't be able to do any output queuing, as the outbound LAN interface is never saturated (the bottleneck is experienced by the DSL modem). Ivan ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/ ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
Re: [c-nsp] Needs some help with QOS
On Tue, March 24, 2009 12:12 pm, Ivan Pepelnjak wrote: What is your upstream connection? If you're using PPPoE, you won't be able to do any output queuing, as the outbound LAN interface is never saturated (the bottleneck is experienced by the DSL modem). If you know what your upstream bandwidth is, you can wrap a shaper around the queueing policy to provide the back-pressure. Useful for all sorts of 'ethernet hand-off' type services where the circuit provider has some other device upstream of your router. Regards, Tim. ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
Re: [c-nsp] Needs some help with QOS
First, thanks to those who pointed out my (should have been obvious) error where I named the access-list qos1 but then tried to reference it with al-qos1. When you're looking for a big problem it's easy to overlook the obvious. On Tue, 2009-03-24 at 12:56 +, Tim Franklin wrote: On Tue, March 24, 2009 12:12 pm, Ivan Pepelnjak wrote: What is your upstream connection? If you're using PPPoE, you won't be able to do any output queuing, as the outbound LAN interface is never saturated (the bottleneck is experienced by the DSL modem). If you know what your upstream bandwidth is, you can wrap a shaper around the queueing policy to provide the back-pressure. Useful for all sorts of 'ethernet hand-off' type services where the circuit provider has some other device upstream of your router. Ok, that also seems to be the point of this link which was provided in another response. http://www.cisco.com/en/US/tech/tk543/tk545/technologies_tech_note09186a00800b2d29.shtml Basically, the virtual interfaces do not implement the back-pressure algorithm necessary to signal that excess packets should be queued by the Layer 3 (L3) queueing system. Ok, so I'm going to have to implement a new solution based on that document. So just a final question, would the solution have worked if it was on a regular interface? I just want to make sure I had the right idea. Regards, - John Lange http://www.johnlange.ca ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
Re: [c-nsp] Needs some help with QOS
http://www.cisco.com/en/US/tech/tk543/tk545/technologies_tech_note0918 6a00800b2d29.shtml Basically, the virtual interfaces do not implement the back-pressure algorithm necessary to signal that excess packets should be queued by the Layer 3 (L3) queueing system. Ok, so I'm going to have to implement a new solution based on that document. So just a final question, would the solution have worked if it was on a regular interface? I just want to make sure I had the right idea. Yes, assuming that your outgoing interface is the bottleneck. For example, if you have a point-to-point uplink, it's usually the bottleneck and the queuing works as expected. But if you have a Fast Ethernet link into the SP network which polices you @ 2 Mbps, the output queue will never form at your output FE interface. Yet again, you'll have to configure shaping to introduce an artificial bottleneck. Ivan http://www.ioshints.info/about http://blog.ioshints.info/ ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
Re: [c-nsp] Needs some help with QOS
Hi. So just a final question, would the solution have worked if it was on a regular interface? I just want to make sure I had the right idea. Yes, in this case the ATM-interface where the PVC lives. But the PVC must be something else than the default ubr class of service. The U in UBR stands for Unspecified, i.e. no QoS. Try vbr-nrt instead. -- Pelle ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
Re: [c-nsp] Needs some help with QOS
On Tue, 2009-03-24 at 13:29 +0100, BALLA Attila wrote: Hi, you should use hierarchical QoS. First of all you should shape the output traffic down to the upstream speed, then you can use the llq inside the shaped class: http://www.cisco.com/en/US/tech/tk543/tk545/technologies_tech_note09186a00800b2d29.shtml I followed the examples on that page but I'm not having any luck. As far as I can tell the queue is dropping at least some packets that it should be prioritizing (look for 582 below). First, here is what i have in my config, and below that is the results of show policy-map interface. As a side question, the file copy now seems to work much differently in that it does a big burst at the start of the copy and then stalls. Is this a burst while the packet queue fills up? --- config --- class-map match-all cm-qos1 match access-group name al-qos1 policy-map parent_shaping class class-default shape average 128000 service-policy child_queueing policy-map child_queueing class cm-qos1 priority percent 70 interface FastEthernet4 service-policy output parent_shaping ip access-list extended al-qos1 permit ip host xxx.xxx.xxx.xxx any permit ip any host xxx.xxx.xxx.xxx - l#show policy-map interface FastEthernet4 Service-policy output: parent_shaping Class-map: class-default (match-any) 157430 packets, 69675635 bytes 5 minute offered rate 15 bps, drop rate 25000 bps Match: any Queueing queue limit 64 packets (queue depth/total drops/no-buffer drops) 47/618/0 (pkts output/bytes output) 15633/18270484 shape (average) cir 128000, bc 512, be 512 target shape rate 128000 Service-policy : child_queueing queue stats for all priority classes: Queueing queue limit 64 packets (queue depth/total drops/no-buffer drops) 0/582/0 (pkts output/bytes output) 3228/2120973 Class-map: cm-qos1 (match-all) 3810 packets, 2978841 bytes 5 minute offered rate 71000 bps, drop rate 25000 bps Match: access-group name al-qos1 Priority: 70% (89 kbps), burst bytes 2200, b/w exceed drops: 582 Class-map: class-default (match-any) 12441 packets, 16201283 bytes 5 minute offered rate 69000 bps, drop rate 0 bps Match: any queue limit 64 packets (queue depth/total drops/no-buffer drops) 46/36/0 (pkts output/bytes output) 12405/16149511 - John Lange http://www.johnlange.ca ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
Re: [c-nsp] Needs some help with QOS
On Tue, 2009-03-24 at 14:39 -0500, John Lange wrote: I followed the examples on that page but I'm not having any luck. As far as I can tell the queue is dropping at least some packets that it should be prioritizing (look for 582 below). ... policy-map parent_shaping class class-default shape average 128000 service-policy child_queueing policy-map child_queueing class cm-qos1 priority percent 70 interface FastEthernet4 service-policy output parent_shaping ... Class-map: cm-qos1 (match-all) 3810 packets, 2978841 bytes 5 minute offered rate 71000 bps, drop rate 25000 bps Match: access-group name al-qos1 Priority: 70% (89 kbps), burst bytes 2200, b/w exceed drops: 582 You only give priority to 89kbps, so if you try to force more traffic through, the excess will of course be dropped when others use the remaining bandwidth. The 5 minute offered rate can never exceed the configured rate, but it can easily land below. Or did you push traffic through for more than five minutes? Maybe the burst allowed in the shaping combined with a harder (less buffered) limit on the WAN interface drops more packets than necessary? As a side question, the file copy now seems to work much differently in that it does a big burst at the start of the copy and then stalls. Is this a burst while the packet queue fills up? Hmm... TCP should take care of finding the correct level. Is the stall permanent, i.e. the connection is dropped? Or does it just take a long break and then resume? You could try playing with the burst size definitions with priority percent 70 burst-size. Default is 200 ms of burst, but that may not suit the purpose. You could try lowering it a little and see if this works better. Regards, Peter ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
Re: [c-nsp] Needs some help with QOS
Hi. Which direction are you trying to prioritize? In the first post the policy were on the Dialer0-interface (traffic from LAN towards DSL), but in the last post it's on the Fa4-interface (traffic from DSL towards LAN). I assume it's the first one because there is less point shaping when going from slow to fast interfaces. It also fit's better with the 128k cap in the shaper. But as Peter sort of pointed out, 70% out of 128k isn't much... -- Pelle ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
Re: [c-nsp] Needs some help with QOS
Hi John, ==match access-group name al-qos1== That acl doesnt exist? Also for DSL, use some appropiate bandwidht values: bandwidth xxx bandwidth receive yyy Use the show policy-map interface dialer 0 to see if the matching works Regards, Wouter 2009/3/23 John Lange j...@johnlange.ca I have crafted and applied some rules which I thought would prioritize traffic from an 871w (via ADSL) to one specific host. The idea is that any traffic destined to this host should be prioritized over all other traffic. Unfortunately my test show absolutely no effect. If I upload a couple of files at the same time, the one with QOS enabled doesn't seem to get any priority. I have two questions: 1) What is wrong with my config? (below) 2) How can I get real-time debugging of the QOS without flooding my console? --- config class-map match-all cm-qos1 match access-group name al-qos1 policy-map pm-qos1 class cm-qos1 priority percent 70 interface Dialer0 bandwidth 200 ip address negotiated ip mtu 1452 ip nat outside ip virtual-reassembly zone-member security out-zone encapsulation ppp dialer pool 1 dialer-group 1 ppp authentication chap callin ppp chap hostname x...@xxx ppp chap password 7 xx service-policy output pm-qos1 ip access-list extended qos1 permit ip host XXX.XXX.XXX.XXX any permit ip any host XXX.XXX.XXX.XXX -- John Lange http://www.johnlange.ca ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/ -- Wouter Prins w...@null0.nl 0x301FA912 ___ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
