in
IEOC CCIE Forums

IEOC - INE's Online Community

Welcome to INE's Online Community - IEOC - a place for CCIE and CCENT candidates to connect, share, and learn. Our Online Community features CCIE forums and discussions for all tracks including Routing & Switching, Voice, Security, Service Provider, Wireless,, and Storage. Through these online communities you can discuss your questions with thousands of your peers, hundreds of CCIE's and INE's own team of world renowned CCIE instructors and authors, Brian Dennis - Quintuple CCIE #2210, Brian McGahan – Triple CCIE #8593, Petr Lapukhov - Quad CCIE #16379, and Mark Snow - Dual CCIE #14073.
Latest post 04-14-2014 7:19 PM by Happy. 13 replies.
Page 1 of 1 (14 items)
Sort Posts: Previous Next
  • 05-06-2013 2:45 AM

    MQC Single-Rate Three-Color Policer

    The solution guide explains how policing burst sizes at one end must be greater than shaping burst sizes at the other end, but the explanation is a little too difficult for me.

    Can someone explain?

    ------------------------------

    By the way, the formula in Step 3 in SG should be corrected as follows (added the words in bold):

     

    if (S <= CBS) then   ### "<=" instead of "<"

      Packet Conforms;

      CBS = CBS - S;

    else if (S < CBS + EBS) then  ### "CBS + EBS" instead of "EBS"

      Packet Exceeds;

      EBS = CBS + EBS - S;  ### "CBS + EBS - S" instead of "EBS - S"

    else

      Packet Violates;

    end if

     

    Filed under: ,
    • Post Points: 35
  • 05-06-2013 3:23 AM In reply to

    Re: MQC Single-Rate Three-Color Policer

    Hi

     

    In production this is the way you would want to do it.

     

    Say you have a link to a service provider. The link can support X bits per second by you only purchase Y. The service provider will police the traffic coming into them at Y  and will either drop excess traffic or mark it so it may be dropped. This is not good, as it may cause your end devices to have to retransmit packets.

     

    To avoid this is good practice to shape your outbound traffic to a rate less than the policing rate enforced by the SP. This way any excess traffic will be queued outbound on you interface rather that potentially dropped  inbound at the provider end.

    • Post Points: 20
  • 05-06-2013 4:07 AM In reply to

    Re: MQC Single-Rate Three-Color Policer

    Hi,

    I can understand your explanation, but why can't we use the "same" shaping value on the CE as the policing value on the PE?  I'm not saying I want to use a shaping value greater than the policing value on the PE.

    SG says not to use a shaping value equal to the policing value but to use a value "smaller than" the policing value at the other end.

    SG seems to be explaining why using a diagram, but I suppose this needs to be explained in more detail.

     

    Shaping value on CE < Policing value on PE   <<<--- Good

    Shaping value on CE > Policing value on PE <<<--- Not good because the PE can drop packets from CE

    Shaping value on CE = Policing value on PE <<<--- SG says Not good but WHY?

    • Post Points: 20
  • 05-29-2013 3:27 AM In reply to

    Re: MQC Single-Rate Three-Color Policer

    Tomohide,

    For any SLA you need to meet the value set or you'll end up losing traffic.  Like the response in the previous post by Robot.

    As you don't know how or what devices, version or code the SP is using the balancing act between your sending rate, Bc and Be and the other end's is a delicate one.  Normally, in real depolyments you'd shape to be better than the SP's receive rate or you run the risk of a mis-match between the two 'Averaging' rates or periods.

    So, in my humble opinion, and from being the SP setting the rates, matching rates is nearly impossible hence not the best option.

    • Post Points: 20
  • 05-29-2013 3:47 AM In reply to

    Re: MQC Single-Rate Three-Color Policer

    "matching rates is nearly impossible hence not the best option."

    I think maybe I can understand that.

    I'll make it a point to always set greater policing values on the PE than the shaping values on the CE to be on the safe side.

    Thanks!

    • Post Points: 20
  • 07-24-2013 4:34 AM In reply to

    Re: MQC Single-Rate Three-Color Policer

    As I understand it it's because shaped traffic is bursty.

    Say you have a line with 100mbit/s and shape it down to 50mbit/s. The resulting traffic flow is not constant flow of 50mbit/s, but a burst of 100mbit/s, followed by an equally long period of inactivity. The lenght of full speed/pause intervals is determined by the shapers Tc, but one can think of it as half a second full speed, then half a second pause, to achieve 50bmit/s.

    On the other side, you have the policer. It affects the traffic flow differently depending on whether the policer's time interval Tc starts with the other side currently sending or pausing.

    To paraphrase out of context: Say you own a stretch of road 500 meters long. You give a gardener the assignment to plant a tree every 10 meters. How many trees will he plant?

    Either 50 or 51, depending on whether he started the stretch with a tree, or with no tree. That extra tree is the additional packet you have to account for, as mentioned in the SG.

     

    Hope this makes sense...

    • Post Points: 35
  • 07-24-2013 6:22 AM In reply to

    Re: MQC Single-Rate Three-Color Policer

    That makes a lot of sense.

    The timing of the policer's and shaper's TC' may not be the same.

    It can happen that the policer's TC interval begin earlier than the shaper's TC interval depending on the timing maybe.

    If the polcing value on the PE is bigger, the PE is able to accomodate the extra packet.

    I think it's clear now.

    Thanks a lot!

    • Post Points: 20
  • 07-24-2013 12:22 PM In reply to

    Re: MQC Single-Rate Three-Color Policer

    I was struggling to understand what is written in the SG. The QOS part of the workbook does not live up to the quality of the rest of the material, IMO. Explanations are lacking and the general feel is of a cut-and-paste reassembly of an earlier version...

    • Post Points: 20
  • 07-24-2013 2:06 PM In reply to

    • JoeM
    • Top 10 Contributor
    • Joined on 04-15-2012
    • Guadalajara, Mexico
    • Elite
    • Points 31,265

    Re: MQC Single-Rate Three-Color Policer

    lukas:
    I was struggling to understand what is written in the SG. The QOS part of the workbook does not live up to the quality of the rest of the material, IMO. Explanations are lacking and the general feel is of a cut-and-paste reassembly of an earlier version...
    I understand what you are saying (difficulty of QoS section), but I kind of want to disagree on the quality of the material (no pun intended).

    The quality is definitely good IMO, but it seems more for someone who already has a foundation in QoS.   It is a real ass-kicker.  Frankly, it is the only section, that makes me think I should give up on the CCIE track. I sometimes study the section, and think maybe I just bit off more than I can chew in this endeavour.

     

    I think the problem, is that QoS was removed completely from the CCNP track, so we come into the CCIE with no knowledge (or forgotten knowledge from previous CCNP track).  Then, there are equations and equations and equations (okay Einstein) and suddendly QoS seems like a blur.

    Again, this is just my opinion, but I think INE doesn't do a super job on ramping-up a little softer on brand new subjects.  I have seen other material where I felt a little more relaxed in learning the subject. 

    But back to the positive part.  I think the INE QoS section is quality material -- especially for coming back and reviewing -- after we finally understand it.

    EDIT:   my current strategy is, after studying QoS, I go back and do labs on other technologies (workbook 2).  This gives a confidence boost about the learning process.  ;-)

    • Post Points: 5
  • 07-24-2013 2:21 PM In reply to

    • JoeM
    • Top 10 Contributor
    • Joined on 04-15-2012
    • Guadalajara, Mexico
    • Elite
    • Points 31,265

    Re: MQC Single-Rate Three-Color Policer

    To be fair (again no pun intended), I want to add that I now have the INE -- Video Learning Series (streaming) -- QoS and Routing Performance.

    This series did help to break it down a little more, and makes it a little easier to understand.  It might be a good investment, for anyone else who is having trouble with this subject.

    • Post Points: 20
  • 07-25-2013 7:46 AM In reply to

    Re: MQC Single-Rate Three-Color Policer

    Thematically, yes, I concur that the teaching is excellent. But still I find the chapter neglected. In practically all cases the change from 12.4(13r)T to 12.4(24)T2 on routers R4,5 and 6 is not reflected, and since the QoS implementation in between those was completely overhauled, most of the output and many configs are simply wrong. Then there are the typos and glitches (eg. 10.25, 10.26, 10.27). Also many references are made to earlier chapters that are not there anymore (or more correctly moved to the back of the chapter).

    Most of these errors are pointed out in the forum in posts as old as 2008 or 2009 (!).

     

    Leaves you scratching your head, but on the good side, stumbling over those prevents you from just glossing over the exercices -> good learning effect...

     

    Lukas

    • Post Points: 20
  • 07-25-2013 3:56 PM In reply to

    • JoeM
    • Top 10 Contributor
    • Joined on 04-15-2012
    • Guadalajara, Mexico
    • Elite
    • Points 31,265

    Re: MQC Single-Rate Three-Color Policer

    I just realized that the task numbering has changed in the QoS section (when they moved legacy to the appendix). Which makes the tasks (and past errors) more difficult to track in the forum.  I also noted your post regarding specific tasks.

    I have created a ticket in the Workbook Feedback form for your post.   Ticket: JVZ-502831

     

     

    • Post Points: 5
  • 07-25-2013 5:43 PM In reply to

    • rriker
    • Top 50 Contributor
    • Joined on 02-21-2013
    • Milwaukee, WI
    • Elite
    • Points 7,030

    Re: MQC Single-Rate Three-Color Policer

    I completely agree with Robot and RelativityDrive.

    I work for an ISP in the reselling of secure internet to banks. My companies BGP routers and DS3/T3 ( We have 2 OC192 connections as well). We use shaping less than what the SLA is, simply since the ISP gives around about bit rate to connect at.

    In the ATC class on the shaping and policing, more specifically Brian McGahan explains the the serialization clock speed can't be exceeded. This is why we need to shape. I'm sure your familiar with the concept. However, since we still use Frame pretty extensively, I'm on a team to migrate 3,700 frame T1s to MPLS over the next 2 years.

    If your Tc and your Bc are configured correctly then you should have no problems forwarding with limited retransmissions.

    If your Bc is less then it's shaped average, this goes in the "bucket" as Be or excess burst, so on a Tc you can add your "bonus burst bits" to the Tc and these bits will be forwarded. They will be above the Bc rate in the time interval of whatever is configured. But since you have bit time bank they won't be affected or dropped.

     

    One other feature of QoS I thought I would struggle with, at least I think I understand it. Is queueing. If the HW queue is full and your VC hold software queue is say 1500. You are able to buffer to the set 1500 if the HW queue is busy, (depends on the configurations on the interface). If the incoming bits can't go anywhere they get dropped. I could go a level deeper if you have multiple queues availble for different QoS policies. but I digress before that.

     

    All in all, the important part is giving your data, whether voice or other wise the best change to go where there being sent. I deal with clients calling in alot that they are seeing dropped traffic and retransmissions. The fun begins there, I have to figure out what type of traffic and so on. Wireshark has become my best friend. Once I determine what the traffic is I can appropriately classify the traffic.

    Another tool I like is NBAR, I am a fan of the PDLM downloads to the flash of the router, makes classification much easier. You can then match protocol if you have NBAR running. I prefer NBAR over IP accounting since it gives you the protocol and top talkers.

     

    HTH

    Rob

    Rob Riker, CCIE #50693, CCNP Service Provider, VCP5-DCV

    • Post Points: 5
  • 04-14-2014 7:19 PM In reply to

    Re: MQC Single-Rate Three-Color Policer

    Thank you for this explanation.  Really helped.

    lukas:

    As I understand it it's because shaped traffic is bursty.

    Say you have a line with 100mbit/s and shape it down to 50mbit/s. The resulting traffic flow is not constant flow of 50mbit/s, but a burst of 100mbit/s, followed by an equally long period of inactivity. The lenght of full speed/pause intervals is determined by the shapers Tc, but one can think of it as half a second full speed, then half a second pause, to achieve 50bmit/s.

    On the other side, you have the policer. It affects the traffic flow differently depending on whether the policer's time interval Tc starts with the other side currently sending or pausing.

    To paraphrase out of context: Say you own a stretch of road 500 meters long. You give a gardener the assignment to plant a tree every 10 meters. How many trees will he plant?

    Either 50 or 51, depending on whether he started the stretch with a tree, or with no tree. That extra tree is the additional packet you have to account for, as mentioned in the SG.

     

    • Post Points: 5
Page 1 of 1 (14 items)
IEOC CCIE Forums Internetwork Expert CCIE Training
About IEOC | Terms of Use | RSS | Privacy Policy
© 2010 Internetwork Expert, Inc. All Rights Reserved