QoS Question!
hurricane1091
Member Posts: 919 ■■■■□□□□□□
in CCIE
Edit - rewritten for clarity
Hello all,
I have been studying QoS - and have an okay understanding of it. However - there is some confusion I have. On the WAN interface when you do a "show interface gigX/X" you will see an input queue line with output drops, and an output queue line.
Input queue: 0/75/0/0 (size/max/drops/flushes); Total output drops: 1775
Queueing strategy: Class-based queueing
Output queue: 0/1000/0 (size/max total/drops)
I was attempting to understand the concept that queuing does not occur until congestion occurs - but can not figure out when that is. I was under the impression it was related to the tx-ring size, but got confused. I am also confused about what the input queue actually is. Output drops mean packets that were destined in the outbound (towards the MPLS cloud) were dropped - but this is listed next to the input queue. Why is this, and what exactly is the input queue? We have an output service policy that is applied on the WAN interface, and basically VOIP traffic is prioritized with 33% of the bandwidth guaranteed, and for the sake of brevity let us say the remaining 67% is the default class. When I issue a "show policy-map interface gigx/x" I can see that the drops occurred in the default class.
My understanding is this - packets come into the router and are sent to the hardware queue of the WAN facing interface. When that fills up (which, I'm unsure of what that limit is), packets begin to go to the software queues until they fill up there as well, and drops occur (again - this is for sure seen in the default class with the show command issued above). Is the input queue filling up, or is it the output queue filling up? One would think input queue is packets coming into the interface from the MPLS cloud, but I believe I have a misunderstanding there.
Hello all,
I have been studying QoS - and have an okay understanding of it. However - there is some confusion I have. On the WAN interface when you do a "show interface gigX/X" you will see an input queue line with output drops, and an output queue line.
Input queue: 0/75/0/0 (size/max/drops/flushes); Total output drops: 1775
Queueing strategy: Class-based queueing
Output queue: 0/1000/0 (size/max total/drops)
I was attempting to understand the concept that queuing does not occur until congestion occurs - but can not figure out when that is. I was under the impression it was related to the tx-ring size, but got confused. I am also confused about what the input queue actually is. Output drops mean packets that were destined in the outbound (towards the MPLS cloud) were dropped - but this is listed next to the input queue. Why is this, and what exactly is the input queue? We have an output service policy that is applied on the WAN interface, and basically VOIP traffic is prioritized with 33% of the bandwidth guaranteed, and for the sake of brevity let us say the remaining 67% is the default class. When I issue a "show policy-map interface gigx/x" I can see that the drops occurred in the default class.
My understanding is this - packets come into the router and are sent to the hardware queue of the WAN facing interface. When that fills up (which, I'm unsure of what that limit is), packets begin to go to the software queues until they fill up there as well, and drops occur (again - this is for sure seen in the default class with the show command issued above). Is the input queue filling up, or is it the output queue filling up? One would think input queue is packets coming into the interface from the MPLS cloud, but I believe I have a misunderstanding there.
Comments
-
hurricane1091
Member Posts: 919 ■■■■□□□□□□
I'll update this since no one ever answered. I really have no idea why output drops are listed next to the input queue, but they are definitely packets that were dropped, but intended to go outbound.
It's definitely the output queue that fills up in the router when packets come into it from the switches. Trouble understanding when that point is, but I've more/less determined it's something most people just leave to defaults unless serious packet loss is occurring for some reason. The odd thing is that the 1 gbps interface will always try to send at line speed, so I am guessing the output queue is always the same size regardless of the circuit (but depending on the router platform). A 10 mbps circuit should have a smaller queue size than a 100 mbps circuit, and I can't really understand how this part all comes together. I suppose it doesn't really matter if the queue size is the same, when you start assigning bandwidth the queues the router knows only that much data can be sent (and these software queues only fill up when congestion occurs), so with that said a lower speed WAN circuit would see the queues fill up faster since they empty at a lower rate, and packet loss is more likely to occur.