We are using the T1 and T2 values on the egress side in a very different way to the way we use T1/T2 on the ingress side.
We are trying to expand our buffers dynamically to prevent the frame from being dropped. How we do this is by not reserving all of our memory per port (in our case we reserved 92%) and contributing to a common pool which can be used for the interfaces that are congested and need the extra buffer space. The dynamic nature of the reserved/max threshold is more flexible that the more regimented method you have described- which may be good for some ports but not others (and you only get two shots since there are only two queue sets). Vik Malhi – CCIE #13890 Managing Partner - IPexpert, Inc. Telephone: +1.810.326.1444 ext 420 Fax: +1.810.454.0130 Mailto: [email protected] On Mar 5, 2012, at 1:44 PM, Kyle Rogers wrote: > Vik, > > Thanks for the explanation, that answered most of my questions and helped > quite a bit. My only other question is why someone would carve out 10% of > the buffers for a queue, but reserve an amount other than 100%. For example, > if I set the Reserved Bandwidth to 80, why wouldn't I just set the buffer > setting to 8 instead? The only explanation I can come up with is that I can > only use whole percentages in the buffer statement and can't put 8.5%, but if > I put 10% buffers and 85% reserved, I can reserve 8.5% of the buffers. Is > that the reason or am I missing a piece of the puzzle? I apologize for > asking so many questions but I'm sort of at an impass in my studies until I > get a firm grasp on this. I will definitely check out the blog. > > Thanks, > Kyle > > On Mon, Mar 5, 2012 at 3:35 PM, Vik Malhi <[email protected]> wrote: > Answers inline. > > For more info please read my 3 part blog on the Catalyst 3750: > http://blog.ipexpert.com/tags/3750-qos/ > > Vik Malhi – CCIE #13890 > Managing Partner - IPexpert, Inc. > > Telephone: +1.810.326.1444 ext 420 > Fax: +1.810.454.0130 > Mailto: [email protected] > > > > > On Mar 5, 2012, at 11:26 AM, Kyle Rogers wrote: > >> QoS is probably the area that I have the most difficulty with - especially >> LAN QoS. I have some general questions. let's use the following sample >> config: >> >> >> mls qos queue-set output 1 buffers 10 10 26 54 >> mls qos queue-set output 1 threshold 2 138 138 92 400 > > You have only showed queue set 1 - we shall assume that the interface is > assigned to queue set 1 but you must check the interface. > >> >> Let's say this is applied to a 100 Mbps interface >> >> So if I understand this correctly: >> >> Queue 1 = 10% of interface bandwidth is reserved (10 Mbps) >> Queue 2 = 10% of interface bandwidth is reserved (10 Mbps) >> Queue 3 = 26% of interface bandwidth is reserved (26 Mbps) >> Queue 3 = 54% of interface bandwidth is reserved (54 Mbps) > > Not really interface bandwidth. When talking about buffer sizes we are > talking about the sizes of the 4 queues = buffer space = memory allocation > per queue. So our buffer size (which is quite small and not published but > potentially 2MB per 4 ports - not important) is for Q1-4 is 10%, 10%, 26%, > 54%. The bandwidth each of the 4 queues has is specified using the srr > commands within the interface. > >> >> In queue 2: >> T1 is set to 138% of bandwidth (138% x 10 Mbps) >> T2 is set to 138% of bandwidth (138% x 10 Mbps) >> T3 is always set to 100% (100% x 10 Mbps) >> Reserved BW = 92% x 10 Mbps >> Maximum Reserved BW = 400% x 10 Mbps > > Let's pretend our buffer per port is 1MB. Q2 has 10% of the buffer which is > 100KB. > > However there is a twist since we are only actually reserving 92% buffers > allocated to Q2. This is defined in the reserved threshold value. So really > what we are reserving or guaranteeing is 92KB of buffer space for Q2. The > remaining 8% goes to what is known as the common pool- which can be used by > anybody (temporarily) as and when it is needed. Q2 is allowed to grab 4x the > buffers if available- so the buffer size could temporarily expand to 4MB > (based on our 1MB per port example). > > So traffic placed into Q2T1 will be dropped when Q2 is 138% full (or when Q2 > has 138KB of it buffers utilized). To get to this value we would have had to > borrow some of the common pool bandwidth since only 92KB is reserved. If > there is no common pool bandwidth then we would have dropped traffic sooner. > > Same for Q2T2. > > Traffic place into Q2T3 will be dropped when Q2 is 400% full (or when Q2 has > 4MB of its buffers utilized). To get to this value we would have had to > borrow a substantial amount of common pool bandwidth. Worst case- we would > drop this traffic when the reserved buffers are full (92KB). > > > >> >> I think the Reserved and Max Reserved are what are tripping me up. >> My questions are: >> >> 1. If I allocated 10% using the "buffers" command and therefore have 10% of >> the interface's Reserved Memory Pool available for Queue 2, why would I then >> cut it down from 10% to 9.2%? >> 2. Does the 400 for Max Reserved mean that T1 +T2 + T3 traffic cannot >> exceed 400% x 10 Mbps of the interface's bandwidth? >> >> I guess I just need a good explanation for the correlation of the numbers in >> the command "mls qos queue-set output 1 threshold 2 138 138 92 400" >> >> Thanks in advance for any assistance. >> >> Kyle >> _______________________________________________ >> For more information regarding industry leading CCIE Lab training, please >> visit www.ipexpert.com >> >> Are you a CCNP or CCIE and looking for a job? Check out >> www.PlatinumPlacement.com > >
_______________________________________________ For more information regarding industry leading CCIE Lab training, please visit www.ipexpert.com Are you a CCNP or CCIE and looking for a job? Check out www.PlatinumPlacement.com
