3 QoS Basic Questions 1

[hunterdw]

Question 1) So, what's the purpose of QoS and when is any service-policy really active? That is, as long as my interface isn't congested, QoS doesn't really accomplish anything, right? It's only when my interface is saturated that any priorities based on class/policy really matters?

Question 2) Consider the following "basic" QoS setup on a router for a CallManager / Cisco IP Phone environment...

class-map match-any voice
match ip dscp ef
match ip precedence 5
class-map match-any sig
match ip precedence 3
match ip dscp af31
match ip dscp cs3

policy-map voip
class voice
priority percent 30
class sig
bandwidth percent 5
class class-default
fair-queue

conf t
int eth0 (outside interface)
service-policy output voip


The "priority percent 30" is initially based on 30% of the interface bandwidth, right? So, if I have a 10BT WIC, QoS assumes my bandwidth priority is 3megs? If I have, 512k instead for my "up-bandwidth", then it's best for me to do one of two things if I really wanted 30% of 512k

a) conf t, int eth0, bandwidth 512 - this way, my 30% is based on 512k instead of 10meg

or

b) explicitly give the priority in kb instead of %... so
policy-map voip
class-map voice
priority 154

Question 3) In my environment, I have mostly 3524XL and 3550 switches. My routers between sites are VPN (1710 at the hub and 1760V at the spokes). How do I ensure correct packet marking end-to-end? CallManager should correctly mark the dscp of my VoIP traffic, right? I just need to tell my router how to prioritize? What about any GBIC uplinks between switches?

I know that's a lot. I'd appreciate any insight...

--DW

 

[rtfmdude]

1 - yes. LLQ only does stuff when you have contention fo' bandwidth.

2 - yes, I think. The only thing that you'll want to check is a 'show interface' to see what the 'available bandwidth' is - I believe 1/4 of interface bandwidth is reserved for fair-queueing, so, it might be best just to use a statis Kb definition. (the 30% that you config vs. the 30% that you get might be different) I think the % is really only useful in situations where you have variable bandwidth interfaces, like, a 4 T1 ppp multilink bundle or something.

3 - remember, callmanager DOES NOT touch the media(which is usually what we're concerned) in most situations. dscp is a layer 3 concept, right? So, your switches should merely switch the traffic(@ L2). No worries there.

hope that helps.

 

[hunterdw]

rtfmdude--

Nice name I did rtfm, but you know as well as I do that it's hard to wade through CCO at times...

Thanks for the insight. You have verified what I suspected.

I appreciate it.

--DW