A QoS dscp question

AF1 through AF4 have the same "priority", they are simply different classes. However, in the case of AFx1 vs. AFx4, AFx4 will have a higher drop probability.

You'd want to perhaps assign your web traffic to a different queue relative to another traffic type.

AF1x = Class 1 AF2x = Class 2 AF3x = Class 3 AF4x = Class 4

x = 1, 2, 3, or 4 and determines the drop probability

snipped-for-privacy@hotmail.com wrote:

Thanks, For six bits DSCP, I know 3 bits for classes, 2 bits for drop probability

st

Well, see, that's the thing. I read the RFC and it seemed a little ambiguous. I have the CiscoPress ONT book, and it's adamant that AF4x vs. AF1x is simply a class distinction. I always thought the same as you that there was indeed a difference. Now, if you use the DSCP class selector (CS1 - 7), then yes, those matter as they'd be processed by non-DSCP compliant nodes.

I'd love to find a clear explanation of the AF classes. As I said, I didn't come away from the RFC with a much better understanding. The Cisco ONT book is the "clearest" one I've seen so far... :-)

AFx4? Perhaps you meant AFx3.

There are only AFx1 through AFx3 (drop precendences).

AF13 has a higher drop precedence than AF12, which has a higher drop precedence that AF11.

Similarly, AF43 has a higher drop precedence than AF42, which has a higher drop precedence that AF41.

Only three levels of drop precedence exist.

There are 64 possible DSCP values (aggregates/classes) which have been divided into 3 DSCP Pools (Codepoint Space). Two of the three pools are experimental. The one that is not, includes four standard PHBs (Per Hop Behaviors).

• Default PHB (best effort)
• Code-Selector PHBs (backward compatibility with IP-precedence)
• Expedited Forwarding PHB (low-loss, low latency, low-jitter, and assured bandwidth)
• Assured Forwarding PHB (with four different forwarding assurances).

You are referring only to the Assured Forwarding PHB which defines four AFx classes, each with three drop precedence values.

AF1x and AF4x are in different "classes". Lose the word "priority".

It has a higher DSCP value (e.g.: AF41 (34), AF11 (10)).

You're switching back and forth between schemes. I recommend that you search for some documents on "DSCP" and "DiffServ", and read them without trying to translate it to prior schemes. After you get a handle on it, then look back in comparison.

There are various models that can be implemented (e.g.: 5 class model, 8 class model, and QoS Baseline Model). The QoS Baseline Model defines up to 11 classes, and is likely to be Cisco's recommended scheme for "Provisioned QoS".

You "classify and mark" packets according to your criteria, and you define the "policy" (policing, scheduling) that you want to apply to the "classes".

Best Regards, News Reader

Try these:

Correct...I meant 3 drop probabilities and 4 classes!

Thanks for the links. I'll take a look at them for myself.

Thanks all. I have checked some cisco papers and get some ideas for it. 1. within a class there are three drop level.(such as af11,af12 and af13) 2. totally no relationsship among classes. (such as af1x,af2x.af3x and af4x)

The paper also give some example, One ISP has three customs and mark one class to each customer (like coustomerA based on transmit rate has af11 (=3D512k). Does ISP has any reason set two more classes to one customer? if yes, any reason?

Second big Question is how to caculate burst value during config QoS?

Thanks, st

Within Assured Forwarding PHB classes ONLY.

Not within Default PHB, Code-Selector PHBs, or Expedited Forwarding PHB classes.

They are classifying and marking on a customer basis, and potentially applying different policies based on those markings. The policies aren't necessarily differentiating the customers based on who's more important.

The ISP could classify specific traffic (e.g.: VoIP) differently than the rest of their traffic with the intent of applying different policies to the resulting classes.

The Cisco QoS Configuration Guide spells that out.

Best Regards, News Reader

Thanks, Checked burst vlaue. It is based on tcp round trip time. usually set 1.5s. so it meant that burst can go 150% of commited speed. right? but i see some config, that bust number is much lower. why? (use different algorithm?)