bandwidth percentages in llq

[Mike Schooley]

we have set up our queues to match our carriers, they have set the top to classes to 40% and 39%, does this really make much of a difference then.  From what I understand, a ratio is calculated and each class gets to transmit according to that ratio, if the bandwith percentages are almost the same the the ratios for those two classes are almost the same so let say in this case the ratio would be 4.0/3.9/2.1.  I'm not sure I see the benefit here in have the 2 differnent classes when they are that close.  Am i understanding this correctly?

policy-map COS_OUT

class COS-EF

   priority 256

class COS-AF41

     bandwidth remaining percent 40

     random-detect dscp-based

class COS-AF31

   bandwidth remaining percent 39

     random-detect dscp-based

class class-default

   bandwidth remaining percent 21

[lgijssel]

The policy performs WRED for two queues; it looks a bit odd but it may be useful.

The behavior of this setup will be quite different from this for example:

class COS-AF41-AF31

   bandwidth remaining percent 79

     random-detect dscp-based

You must check the class-maps to see what exactly is matched.

Admittedly it seems very straightforward but having a class COS-AF41 does not say anything about what is classified as such. Its just a name.

regards,

Leo

[Mike Schooley]

you say that the behaviour is substantially different, why, lets say it was 40/40/20 or 80/20 for easy math.  So ratios would be calculated and 4 packets of af41 the 4 packets of af31 then 2 packets of be, vs 8 packets af41_31, 2 packets be.  I suppose thats substantially different sort of? is af41 af31 af41 af31 af41 af31af41 af31 be be subtantially different that af41 af41 af41 af41 af31 af31 af31 af31 be, I guess it does guarantee af41 as it could be af31 af31 af31 af31 af31 af31 af31 af31 af31 be be as there would be no differentiation between classes.

[lgijssel]

Mike Schooley wrote:
 you say that the behaviour is substantially different, why, lets say it was 40/40/20 or 80/20 for easy math.  So ratios would be calculated and 4 packets of af41 the 4 packets of af31 then 2 packets of be, vs 8 packets af41_31, 2 packets be.  I suppose thats substantially different sort of? is af41 af31 af41 af31 af41 af31af41 af31 be be subtantially different that af41 af41 af41 af41 af31 af31 af31 af31 be, I guess it does guarantee af41 as it could be af31 af31 af31 af31 af31 af31 af31 af31 af31 be be as there would be no differentiation between classes.

No Mike, the diference I mean is that one class may map just match af41 while the other one does af31-33 plus cs3 for example.

Obviously af41 would get a much bigger share of the overal bw.

[jmcgrady1]

---

@lgijssel wrote:

class COS-AF41-AF31

   bandwidth remaining percent 79

     random-detect dscp-based

You must check the class-maps to see what exactly is matched.

Admittedly it seems very straightforward but having a class COS-AF41 does not say anything about what is classified as such. Its just a name.

---

What might the class map look like for this example?  How does the classifier know the traffic is af41 - af31 ?

[Joseph W. Doherty · Accepted solution]

Disclaimer

The    Author of this posting offers the information contained within this    posting without consideration and with the reader's understanding that    there's no implied or expressed suitability or fitness for any  purpose.   Information provided is for informational purposes only and  should not   be construed as rendering professional advice of any kind.  Usage of  this  posting's information is solely at reader's own risk.

Liability Disclaimer

In    no event shall Author be liable for any damages whatsoever  (including,   without limitation, damages for loss of use, data or  profit) arising  out  of the use or inability to use the posting's  information even if  Author  has been advised of the possibility of such  damage.

Posting

If your asking what's the difference between something like

class COS-AF41

     bandwidth remaining percent 40

     random-detect dscp-based

class COS-AF31

   bandwidth remaining percent 39

     random-detect dscp-based

vs.

class COS-AF41_n_AF31

     bandwidth remaining percent 79

     random-detect dscp-based

The former would dequeue from two queues while the latter would dequeue from a single queue.  The dual queues would likely better interleave the two classes of traffic.  (NB: if all three classes, above, also enabled FQ, dequeuing would be very similar.)

The other difference, since you've enabled WRED, behavior will be different because defaults settings will apply WRED against two queues instead of one queue (even though defaults will be a little different for each DSCP marking).  For example, each class queue might have 20 packets, and WRED might not engage at all, but if the combined queue had 40 packets, WRED might start dropping packets.  Or say, one queue had 10 packets and the other 40, the former might not engage WRED but the latter might, but if you combine them, 50 packets in one queue might subject the 10 packets of the one class to drops.  (NB: Although WRED has settings per marking, it counts against total queue depth, not just packets of that class marking.)