08-15-2007 08:45 AM - edited 03-03-2019 06:20 PM
Hi, I have a situation where we are paying for 100mb MAN connection. Our uplink to the MAN is a gigabit connection. So I'm guessing they will police our traffic to 100mb.
Voice will be running over this link along with SAN traffic, so we will be applying the below QoS policy:
class-map match-any SAN-TRAFFIC
match access-group name SAN-Traffic
class-map match-all VOICE
match ip dscp ef
class-map match-any CALL-SIGNALING
match ip dscp cs3
match ip dscp af31
!
!
policy-map DC-DC-EDGE
class VOICE
priority percent 5
class CALL-SIGNALING
bandwidth percent 5
class SAN-TRAFFIC
bandwidth percent 60
class class-default
bandwidth percent 5
ip access-list extended SAN-Traffic
permit ip host 10.8.5.10 host 10.0.5.10
Is it recommended that I police the traffic before it gets to the Provider or should I just stick with my percentages and let the provider police the traffic?
Thanks,
Brad
08-15-2007 08:51 AM
Let your provider police it for you. . .just remember to tell the WAN interface that it only has 100Mb to work with, not the Gb it auto-negotiated, that way your queues will get used.
08-15-2007 09:01 AM
So if set the bandwidth on the interface to 100000 - it will know that 5% for Voice is actually 5 meg and not 50 meg?
Thanks,
Brad
08-16-2007 07:45 AM
Correct about the bandwidth. Think of it as creating a false congestion so the router knows about it.
And as was stated before, QoS on this link might not even really be needed. I don't think it would HURT to implement it, just in case, but you really should baseline your connection (possibly with AutoQoS) and go from there.
08-17-2007 06:35 AM
Adam does touch upon an interesting point, if you can (physically) run the WAN interface at 100 Mbps, you wouldn't need to use a shaper. However, using the shaper you can easily contract with the provider to increase your effective bandwidth at any value between 100 Mbps and gig.
08-15-2007 08:53 AM
Hi Brad,
Suggest that you throughly test the circuit for voice in presence of intensive data traffic.
You may find that considering the speed you have purchased there is no need for any QoS setting whatsoever.
That would be the best situation, otherwise you may need to shape (that is better than policing) traffic at the actual speed contracted with the SP.
Hope this helps, please rate post if it does!
08-15-2007 10:13 AM
If the provider is actually policing your traffic, then it's fine not do anything if you don't mind them discarding your traffic when you go above your contracted rate or if you know that your traffic would never go above your contracted rate. An example of the latter would be a front side 100 Mbps (LAN) connection with your backside gig (MAN) connection that's policed to 100 Mbps.
If it's possible that you can oversubscribe the 100 Mbps MAN, and the oversubscription can be due to the SAN traffic, and considering you want to protect your voice, then QoS is a must.
If you only have voice and SAN, you only need to provide the two classes for voice (as you done); also assuming you've allocated enough bandwidth. You can leave off SAN-TRAFFIC, i.e. let it fall into the default.
If you might have other traffic other than voice and SAN, you can go with a policy much like what you have except you can set the SAN bandwidth to a setting to split the bandwidth with your other traffic as you like. The 60% setting doesn't limit SAN to 60% it means it gets 60% if the other 40% is in demand. (Except for LLQ, bandwidth is a floor, not a maximum.) If this is what you intend, that?s find.
However, your policy would assume all the bandwidth of the gig is available so you need to create a hierarchical policy that limits the total to 100 Mbps.
e.g. (12.4 syntax - assumes gig limited to 100)
policy-map DC-DC-EDGE-shape
class class-default
shape average percent 10
service-policy DC-DC-EDGE
08-15-2007 12:27 PM
Hi,
Won't the shaping add a lot of delay which will impact voice? (Using a 2851 router).
08-15-2007 12:41 PM
hi,
not at all. Voice traffic goes into priority queue, so chances are that is never shaped.
True is that with a nested policy-map, one or two voice packets it could be subject to shaping if the gap with the packet that has been transmitted immediately before is small.
But even if was shaped, delay at speed of 100 mbps is of a sub-millisecond order of magnitude, so never an issue.
Shaping is better than policing because it allow you to not drop traffic, still respecting a traffic contract.
As a recognition to those providing answers, please rate useful posts!
Discover and save your favorite ideas. Come back to expert answers, step-by-step guides, recent topics, and more.
New here? Get started with these tips. How to use Community New member guide