Cisco priority queuing help

Posted on 2006-10-26
Last Modified: 2008-01-09

When configuring Priority queuing for QOS I know that if high priorty packets comes through there is a chance that lower priorty packets will be dropped. My question is what happens if my high priority queue is full and more high priority comes in? Would the packets be  dropped?

Question by:iamuser
  • 3
  • 3
LVL 12

Accepted Solution

pjtemplin earned 150 total points
ID: 17815374
If they arrive at a sustained rate faster than the PQ can drain (egress interface rate) or is allowed to drain (setting the PQ to a specific bandwidth, which is actually an often-used setting), yes, packets will be dropped.

Expert Comment

ID: 17827725

"The key to understanding latency is to understand the concept of
micro-congestion:  It doesn't matter what the overall utilization is on
the network:  Quoting numbers like "30%" even when measured with any of
the numerous tools, is unlikely to yield the issues that contribute so
heavily to latency overall.  If 2 packets arrive at the input queue at
the same time, even if the output queue is empty, there's gonna be a
collision.  If you can enforce some queuing priority, then the higher
priority packet(s) get pushed through first.  Ask yourself, now, what
happens when 2 high priority packets arrive at the input queue at the
same time?  Uhoh. micro-congestion, again."

Two high priority packets arriving at the same time = dropped packets.  However, retransmition is built into the protocols, so unless you saturate the device continually, it's just part of the game... it is expected to happen.

Best Wishes,

Jeffery Smith
LVL 12

Expert Comment

ID: 17828528
Two high priority packets arriving at the same time != dropped packets.

It only means one of them is queued for the time it takes the other to go out the wire.  Big deal.  On a T1, a 1500-byte packet (as large as can be on your Ethernet segments) takes 8ms to transmit.  8ms is defintely not 0ms, but it's not significant in the scope of VoIP latency budgets, and if it's VoIP, it's a much smaller packet so we're talking more like 0.5ms.
Windows Server 2016: All you need to know

Learn about Hyper-V features that increase functionality and usability of Microsoft Windows Server 2016. Also, throughout this eBook, you’ll find some basic PowerShell examples that will help you leverage the scripts in your environments!


Assisted Solution

Jeffesmi earned 150 total points
ID: 17829787
Sorry, I was a bit off topic talking about collisions.  My assumption was a single interface, where pjtemplin is obviously talking about packets coming from multiple interfaces.  As two packets at the same time, on the same interface, definately does mean a colision which means packet loss.  More on point, yes, if high priority packets come in while the high-priority queue is full, packets would be dropped.  This is because the Cisco model services the high priority queue first.  If the queue is full, there is nowhere else to go.  If you are wondering if a low priority packet would be dropped to allow room for a high priority packet, I don't think it would as you define queue size and rules to determine what is high, medium, and low priority. I believe that Cisco uses tail-drop, but that may be configurable. I couldn't find any information on this specific area, but here is some interesting reading:

Priority queuing-This technique uses multiple queues, but queues are serviced with different levels of priority, with the highest priority queues being serviced first. Figure Q-2 illustrates Cisco's priority queuing scheme. When congestion occurs, packets are dropped from lower-priority queues. The only problem with this method is that lower-priority queues may not get serviced at all if high-priority traffic is excessive. Packets are classified and placed into queues according to information in the packets. For example, Cisco routers can be programmed to prioritize traffic for a particular port into high-, medium-, or low-priority queues. Priority schemes may be abused by users or applications that mark packets with priorities that are not allowed. Admission control functions can monitor this. See "Admission Control."

Packets may arrive at queues in bursts from multiple devices, and a device may temporarily receive more packets than it can process. Buffers hold packets until a device can catch up. If the device cannot catch up, buffers fill up and new incoming packets are dropped. This is called "tail drop."

Two queue disciplines can be used in place of tail drop. "Random drop on full" drops randomly selected packets. "Drop front on full" drops packets at the front of queues. Dropping on the front is used to more quickly notify a sender of congestion. Consider a case where a sender has filled a queue and tail drop occurs. The receiver will continue to receive packets and not know that congestion is occurring until it notices the dropped packet. Dropping at the front tells the receiver right away to signal congestion (by failing to acknowledge).

Best Wishes,

Jeffery Smith
LVL 12

Expert Comment

ID: 17830677
Dude, the high priority queue is a QUEUE, not a single packet.  If two packets arrive, that does not mean the second one gets dropped.  

Example: T1 interface on a Cisco 7507, configured as follows:

class-map match-all voip
  match ip precedence 5
policy-map voip-qos
  class voip
    priority percent 50
  class class-default
interface Serial0/1/1/14:1
 ip address
 no ip redirects
 no ip directed-broadcast
 no ip proxy-arp
 encapsulation ppp
 no fair-queue
 service-policy output voip-qos

dist1-hstn#sh pol int s0/1/1/14:1 out

  Service-policy output: voip-qos (1035)

    queue stats for all priority classes:

      queue limit 192 (packets)
      (queue depth/total drops/no-buffer drops) 0/0/0
      (pkts queued/bytes queued) 97/41341

    Class-map: voip (match-all) (1036/1)
      97 packets, 41341 bytes
      5 minute offered rate 0 bps, drop rate 0 bps
      Match: ip precedence 5  (1037)
      Priority 50 (%) (768 kbps) burst 19200 (bytes)

    Class-map: class-default (match-any) (1039/0)
      108882061 packets, 63161272536 bytes
      5 minute offered rate 2000 bps, drop rate 0 bps
      Match: any  (1040)
        108882061 packets, 63161271110 bytes
        5 minute rate 2000 bps
      queue limit 192 (packets)
      (queue depth/total drops/no-buffer drops/flowdrops) 0/25831/0/25831
      (pkts queued/bytes queued) 110293381/63204190043
      Fair-queue: per-flow queue limit 48

Here you'll see that a priority queue configured to allow 768kbps (50% of a T1) has a queue size of 192 packets.  So yes, if 191 packets have been queued up awaiting output scheduling and 2 more arrive before 1 of those 191 can be scheduled, 1 will be dropped.  But it would take a lot of work to have 190+ packets arrive in no more than 16ms (the time it'd take to transmit a 1500-byte packet at 768kbps), and if priority packets were arriving at a steady rate of more than 768kbps, then this customer probably needs to rethink their bandwidth needs.

Expert Comment

ID: 17831770
  Doesn't a T1 primarily work over ATM?  If so, can two packets arrive at the same time? I believe that I understand what you are saying. You are talking about the queue.  As I said, I was off point talking about collisions.  Bottom line is that the statement is correct.  If two packets arrive at the interface at the same time, the packets are dropped. <-- This was off topic as I've said. We aren't talking about the interface.  We are talking about priority queues.>  However, in non-CSMA/CD-CA systems, this shouldn't/can't happen.  If I'm wrong about this, I have to go back an study my network books again as I'm 99.99999% sure that a single interface can't accept two packets and packets can't get to the queue until the interface accepts them.  YOUR POINT IS VERY VALID.  The queue was being discussed. I don't know why my mind pulled out interface information other than early onset brain rot.  

I don't know anything about the network being discussed.  They could have an OC-24, multiple load balanced T3s, maybe they have 10,000 phones running over VOIP and have a horrendous amount of high priority information.  Maybe it was just a question from someone studying for their Cisco test.  I don't know, but if there were no situations where priority queuing was appropriate, I don't think Cisco would have wasted their time on it.  I just like answering questions.  It makes me feel all warm and tingly. :-)

Best Wishes,

Jeffery Smith

Featured Post

Live: Real-Time Solutions, Start Here

Receive instant 1:1 support from technology experts, using our real-time conversation and whiteboard interface. Your first 5 minutes are always free.

Question has a verified solution.

If you are experiencing a similar issue, please ask a related question

Suggested Solutions

Cable Modem Provisioning from DPoE compliant server  This Article is to support CMTS administrators to provide an overview of DOCSIS compliance configuration file, and to provision a cable modem located at customer place from a Back office serve…
Sometimes you have to pull out old tricks to get a new firewall to work… While we were installing a new Sonicwall at a customers site we found that sites they were able to visit before were not working.  It seemed random and we could not understa…
In a recent question ( here at Experts Exchange, a member asked how to add page numbers to a PDF file using Adobe Acrobat XI Pro. This short video Micro Tutorial sh…
A short tutorial showing how to set up an email signature in Outlook on the Web (previously known as OWA). For free email signatures designs, visit If you want to manage em…

786 members asked questions and received personalized solutions in the past 7 days.

Join the community of 500,000 technology professionals and ask your questions.

Join & Ask a Question