WEBVTT

0:00:02.900000 --> 0:00:07.480000
 Hello and welcome to this video titled
 congestion management control with

0:00:07.480000 --> 0:00:09.740000
 queuing and scheduling.

0:00:09.740000 --> 0:00:15.220000
 In this video I'm going to define what
 queuing means and what scheduling

0:00:15.220000 --> 0:00:20.220000
 means. We're going to talk about what
 is congestion management and why

0:00:20.220000 --> 0:00:25.100000
 do we need it. And I'm just going to
 briefly show you some queuing and

0:00:25.100000 --> 0:00:28.700000
 scheduling features just so you're familiar
 with the terminology of those

0:00:28.700000 --> 0:00:33.500000
 features. So let's
 start with queuing.

0:00:33.500000 --> 0:00:36.380000
 Let's define what queuing means.

0:00:36.380000 --> 0:00:40.720000
 So in order to do this, let's
 just look at something here.

0:00:40.720000 --> 0:00:47.300000
 So here is our egress interface.

0:00:47.300000 --> 0:00:50.080000
 It's about to transmit
 stuff on the wire.

0:00:50.080000 --> 0:00:57.560000
 So, and here this represents
 our memory buffers.

0:00:57.560000 --> 0:01:05.580000
 And we know those memory buffers can
 be used by a whole bunch of things

0:01:05.580000 --> 0:01:09.660000
 including ARP tables,
 routing tables.

0:01:09.660000 --> 0:01:13.320000
 Basically it's whatever memory needs
 to be used in this router switch

0:01:13.320000 --> 0:01:14.860000
 is used for that.

0:01:14.860000 --> 0:01:25.300000
 Now, with these platforms we have
 things called transmit rings.

0:01:25.300000 --> 0:01:33.700000
 What a transmit ring is is a certain
 preallocated number of buffers like

0:01:33.700000 --> 0:01:39.520000
 maybe these ones right here.

0:01:39.520000 --> 0:01:44.440000
 And as frames come in inbound or ingress
 interfaces and they get looked

0:01:44.440000 --> 0:01:48.660000
 up and they have to go out this interface,
 let's say this is fast ethernet

0:01:48.660000 --> 0:01:54.600000
 0 slash 1, they will be placed into
 these memory buffers here which are

0:01:54.600000 --> 0:01:56.140000
 called the transmit ring.

0:01:56.140000 --> 0:02:01.700000
 Now, if we never fill these up, okay,
 if this is as far as it gets and

0:02:01.700000 --> 0:02:07.260000
 we don't go outside of that sort of
 purplish paint box right there, that

0:02:07.260000 --> 0:02:10.380000
 means we can transmit
 at line rate.

0:02:10.380000 --> 0:02:14.960000
 So any, as long as we don't fill up that
 box, we can go up to 100 million

0:02:14.960000 --> 0:02:16.020000
 bits per second.

0:02:16.020000 --> 0:02:19.620000
 So clearly there would be a lot more
 boxes that I've defined right here.

0:02:19.620000 --> 0:02:23.740000
 But if packets are always allowed to
 go onto the transmit ring and they

0:02:23.740000 --> 0:02:28.060000
 never exceed the transmit ring,
 there is no congestion.

0:02:28.060000 --> 0:02:30.180000
 We're transmitting at line rate.

0:02:30.180000 --> 0:02:35.760000
 Where problems occur is where packets
 are coming in so fast from our ingress

0:02:35.760000 --> 0:02:40.820000
 interfaces that they're exceeding the amount
 of buffer space we have available

0:02:40.820000 --> 0:02:44.840000
 or that we've allocated
 for the transmit ring.

0:02:44.840000 --> 0:02:50.680000
 So in that particular case, then the
 CPU of the switch might say, okay,

0:02:50.680000 --> 0:02:58.100000
 well, all of, or the router, router
 or switch, it might say all of these

0:02:58.100000 --> 0:02:59.060000
 boxes are available.

0:02:59.060000 --> 0:03:03.520000
 So this is here within this black square
 allocated to fast ethernet zero

0:03:03.520000 --> 0:03:08.920000
 slash one. The ones that are in
 purple, that's my transmit ring.

0:03:08.920000 --> 0:03:14.500000
 But if I go beyond that, I can
 still store stuff in here.

0:03:14.500000 --> 0:03:22.040000
 Now, without quality of service, without
 QoS turned on, it's first in,

0:03:22.040000 --> 0:03:24.560000
 first out. There's no priority.

0:03:24.560000 --> 0:03:27.320000
 There's no way of differentiating
 the traffic.

0:03:27.320000 --> 0:03:29.520000
 It's just first in, first out.

0:03:29.520000 --> 0:03:33.640000
 This is a situation where we would
 want to use quality of service.

0:03:33.640000 --> 0:03:39.160000
 Because by turning quality of service
 on, it doesn't affect how many of

0:03:39.160000 --> 0:03:42.600000
 these boxes are allocated
 to that interface.

0:03:42.600000 --> 0:03:45.680000
 But now we have control
 over those boxes.

0:03:45.680000 --> 0:03:50.880000
 We as a network engineer or admin can
 say, okay, with quality of service,

0:03:50.880000 --> 0:03:56.840000
 instead of looking to actually do anything
 about, now I can say, all right,

0:03:56.840000 --> 0:04:02.580000
 I want these buffers right here, I want
 to have control over those, and

0:04:02.580000 --> 0:04:10.120000
 that'll be my low priority queue.

0:04:10.120000 --> 0:04:23.820000
 And I want maybe these ones right
 here to be my high priority queue.

0:04:23.820000 --> 0:04:29.900000
 So, queuing is the act of doing this,
 of taking our memory buffers and

0:04:29.900000 --> 0:04:34.260000
 giving us some control over it.

0:04:34.260000 --> 0:04:41.260000
 So once we enable QoS, we start
 dealing with queuing.

0:04:41.260000 --> 0:04:47.720000
 As this mentions, a single egress interface
 may have multiple associated

0:04:47.720000 --> 0:04:51.120000
 queues differentiated by priority.


0:04:51.120000 --> 0:04:57.600000
 Now, when we're talking about switching,
 switches, at least Cisco switches,

0:04:57.600000 --> 0:05:00.340000
 most of what they do, especially when
 it comes to the world of quality

0:05:00.340000 --> 0:05:03.640000
 of service, is done in hardware.

0:05:03.640000 --> 0:05:05.320000
 Why is that relevant?

0:05:05.320000 --> 0:05:11.240000
 Well, because on a switch, if I turn on
 QoS globally with a command, there's

0:05:11.240000 --> 0:05:15.460000
 no way that depending up on what the
 module or the port is that's in that

0:05:15.460000 --> 0:05:19.960000
 switch, it will have a predefined
 quantity of queues.

0:05:19.960000 --> 0:05:23.900000
 In other words, I can't say, one thing
 I cannot do is I can't say, okay,

0:05:23.900000 --> 0:05:28.880000
 you interface right here, I want to
 give you four queues, low, medium,

0:05:28.880000 --> 0:05:30.100000
 high, medium, high.

0:05:30.100000 --> 0:05:32.480000
 You interface over here, I'll
 just give you two queues.

0:05:32.480000 --> 0:05:34.280000
 You will go over here, I'll
 give you eight queues.

0:05:34.280000 --> 0:05:35.380000
 You can't do that.

0:05:35.380000 --> 0:05:39.400000
 On a switch, because everything's done
 in hardware, that particular module

0:05:39.400000 --> 0:05:42.520000
 will always come predefined with a
 certain amount of queues the moment

0:05:42.520000 --> 0:05:44.420000
 you turn on QoS.

0:05:44.420000 --> 0:05:47.800000
 So, this would be actually part of your
 buying, your purchasing decision,

0:05:47.800000 --> 0:05:51.080000
 is if you know you're going to have
 to do QoS in your network, you might

0:05:51.080000 --> 0:05:54.100000
 want to say, okay, I've got these two
 modules right here I could buy.

0:05:54.100000 --> 0:05:57.160000
 They're both fast-heath
 and at line cards.

0:05:57.160000 --> 0:06:02.100000
 But maybe this one right here, when
 I turn on QoS gives me four queues

0:06:02.100000 --> 0:06:07.760000
 per port. This one over here
 gives me two queues per port.

0:06:07.760000 --> 0:06:11.060000
 Well, the one with four queues is going
 to give you more flexibility,

0:06:11.060000 --> 0:06:13.640000
 more things you can do within
 those four queues.

0:06:13.640000 --> 0:06:15.820000
 So, that might be part of
 your buying decision.

0:06:15.820000 --> 0:06:20.240000
 Now, if we're talking about a router,
 typically with routers, the queues

0:06:20.240000 --> 0:06:22.060000
 are software-driven queues.

0:06:22.060000 --> 0:06:26.040000
 They're created dynamically by the CPU
 and you have a lot more control.

0:06:26.040000 --> 0:06:29.160000
 You actually could say something like,
 I want this port to have four queues

0:06:29.160000 --> 0:06:30.840000
 and this port to have seven.

0:06:30.840000 --> 0:06:34.060000
 I want this queuing here
 to be based on DSCP.

0:06:34.060000 --> 0:06:37.260000
 I want this queuing method over
 here based on IP precedence.

0:06:37.260000 --> 0:06:39.080000
 You have a lot more flexibility.

0:06:39.080000 --> 0:06:42.800000
 So, that's sort of in summary what
 this bullet point is saying is that

0:06:42.800000 --> 0:06:46.980000
 an interface could have one queue,
 two queues, eight queues.

0:06:46.980000 --> 0:06:49.660000
 It all depends on is it
 a router or switch?

0:06:49.660000 --> 0:06:51.060000
 What have you configured?

0:06:51.060000 --> 0:06:54.000000
 What have you not configured?

0:06:54.000000 --> 0:06:59.380000
 So, when we talk about QoS queuing features,
 we're talking about features

0:06:59.380000 --> 0:07:02.680000
 that are designed to give you
 control over those queues.

0:07:02.680000 --> 0:07:06.480000
 If it's a router, it'll give you control
 over how many queues you can

0:07:06.480000 --> 0:07:10.960000
 have and how to direct frames
 into those queues.

0:07:10.960000 --> 0:07:14.260000
 In other words, do you want your low
 priority queue, you know, what's

0:07:14.260000 --> 0:07:15.520000
 going to go in there?

0:07:15.520000 --> 0:07:17.340000
 You get to choose that.

0:07:17.340000 --> 0:07:20.620000
 You can say, okay, well, certain IP
 precedence values will go in there

0:07:20.620000 --> 0:07:25.460000
 or certain DSCP values will go in there
 or maybe it's based on applications.

0:07:25.460000 --> 0:07:28.260000
 Maybe HTTP traffic
 will go in there.

0:07:28.260000 --> 0:07:31.600000
 QoS features give you
 that flexibility.

0:07:31.600000 --> 0:07:35.800000
 Some QoS features when it comes to
 queuing also allow you to configure

0:07:35.800000 --> 0:07:40.340000
 preemptive drops of traffic to always
 make sure that there's room in that

0:07:40.340000 --> 0:07:43.100000
 queue for the higher
 priority stuff.

0:07:43.100000 --> 0:07:45.220000
 So, that's what queuing is.

0:07:45.220000 --> 0:07:49.980000
 So, when we talk about a QoS feature that
 is a queuing feature, it's either

0:07:49.980000 --> 0:07:55.140000
 creating the quantity of queues I want
 per port, if one of my router,

0:07:55.140000 --> 0:08:02.460000
 I can do that, and or determining what
 traffic will go into those queues.

0:08:02.460000 --> 0:08:06.780000
 That's queuing. All right, so let's
 imagine for a moment we've already

0:08:06.780000 --> 0:08:09.680000
 done that. So, we've already figured,
 okay, for this one interface right

0:08:09.680000 --> 0:08:13.620000
 here, I've already determined there's
 going to be four queues, low, medium,

0:08:13.620000 --> 0:08:17.840000
 high, critical, and I've already determined
 that certain IP precedence

0:08:17.840000 --> 0:08:20.840000
 or DSCP values will go
 into each one of those.

0:08:20.840000 --> 0:08:22.160000
 I'm done with that.

0:08:22.160000 --> 0:08:25.960000
 Okay, now, imagine for a moment that
 each one of those queues has stuff

0:08:25.960000 --> 0:08:29.540000
 in it. Each one of those queues
 is fairly full of things.

0:08:29.540000 --> 0:08:32.580000
 So, the next thing we have
 to determine is scheduling.

0:08:32.580000 --> 0:08:34.660000
 So, what is the scheduler?

0:08:34.660000 --> 0:08:39.040000
 The scheduler is the feature responsible
 for pulling packets out of those

0:08:39.040000 --> 0:08:42.540000
 queues and putting
 them on the wire.

0:08:42.540000 --> 0:08:48.680000
 So, QoS scheduling features modify
 or affect that behavior.

0:08:48.680000 --> 0:08:52.440000
 So, some QoS scheduling features would
 give you the ability to say, okay,

0:08:52.440000 --> 0:08:57.940000
 on the low priority queue, I want to
 service that for like 15% of the

0:08:57.940000 --> 0:09:02.580000
 total time, then pause, move on to
 the medium priority queue for maybe

0:09:02.580000 --> 0:09:04.740000
 another 20% of the time.

0:09:04.740000 --> 0:09:09.660000
 Pause, or you might, another scheduling
 feature might say, okay, well,

0:09:09.660000 --> 0:09:14.380000
 I'm going to pull 20 packets out
 of the low priority queue.

0:09:14.380000 --> 0:09:17.640000
 Then I'm done, if there's some left
 in there, I'm going to, I'm done,

0:09:17.640000 --> 0:09:23.280000
 I'm going to move 20 packets out
 of the low priority queue.

0:09:23.280000 --> 0:09:27.320000
 So, you can see scheduling is all
 about how do I service the queue?

0:09:27.320000 --> 0:09:31.960000
 How do I deal with what's in there right
 now and get it on the wire and

0:09:31.960000 --> 0:09:36.940000
 how do I equally spend my time among
 the various queues that exist?

0:09:36.940000 --> 0:09:42.540000
 Now, on switches, queuing and scheduling,
 a lot of times are completely

0:09:42.540000 --> 0:09:43.740000
 separate features.

0:09:43.740000 --> 0:09:47.260000
 You can figure one feature to do the
 queuing part of it, another feature

0:09:47.260000 --> 0:09:50.220000
 to do the scheduling
 part of it, okay.

0:09:50.220000 --> 0:09:54.440000
 Other times on switches and routers,
 one feature will do it all.

0:09:54.440000 --> 0:09:57.860000
 Queuing and scheduling will be built
 into the feature, and on one line

0:09:57.860000 --> 0:10:02.560000
 of command, you'll have to do both your
 queuing and your scheduling stuff.

0:10:02.560000 --> 0:10:07.180000
 So, if you've ever learned about traffic
 shaping, either in this video

0:10:07.180000 --> 0:10:11.580000
 series or another video series or just
 through reading, traffic shaping

0:10:11.580000 --> 0:10:15.120000
 would be an example of a scheduling
 function, right?

0:10:15.120000 --> 0:10:20.040000
 Traffic shaping does not dictate
 how many queues an interface has.

0:10:20.040000 --> 0:10:25.440000
 Traffic shaping does not dictate what
 type of traffic goes into a particular

0:10:25.440000 --> 0:10:29.860000
 queue. Traffic shaping deals with a
 queue that's already got stuff in

0:10:29.860000 --> 0:10:34.520000
 it and how fast we're going to play
 it out onto the wire and cap it at

0:10:34.520000 --> 0:10:38.340000
 a certain rate. So, traffic shaping
 would be a scheduling function of

0:10:38.340000 --> 0:10:43.620000
 queuing. So, we had to first sort of define
 the differences between queuing

0:10:43.620000 --> 0:10:48.300000
 and scheduling before we talk about congestion
 management at a high level,

0:10:48.300000 --> 0:10:51.700000
 because this wouldn't make much sense
 to you if you didn't know what queuing

0:10:51.700000 --> 0:10:53.280000
 and scheduling are.

0:10:53.280000 --> 0:10:57.240000
 So, congestion management is all about
 how do we deal with those queues?

0:10:57.240000 --> 0:11:01.740000
 Because remember, if I'm not congested,
 everything's going on my transmit

0:11:01.740000 --> 0:11:06.260000
 ring. Those few memory cells are reserved
 for my transmit ring, everything's

0:11:06.260000 --> 0:11:10.240000
 going in there, and I don't need
 any additional buffer space.

0:11:10.240000 --> 0:11:13.040000
 So, congestion management is dealing
 with, okay, when I do need additional

0:11:13.040000 --> 0:11:17.360000
 buffer space, let me get
 some control over that.

0:11:17.360000 --> 0:11:21.880000
 So, various features we have could help
 us to define how many queues we're

0:11:21.880000 --> 0:11:26.240000
 going to have, what packets are going
 to go into those queues based on

0:11:26.240000 --> 0:11:31.560000
 classification, maybe selectively dropping
 them once they get in there

0:11:31.560000 --> 0:11:35.920000
 to leave room for higher priority traffic,
 and the scheduling of those

0:11:35.920000 --> 0:11:42.740000
 packets. So, why do we need
 congestion management?

0:11:42.740000 --> 0:11:43.860000
 Why do we need these features?

0:11:43.860000 --> 0:11:47.240000
 Well, because by default, like I drew
 on the whiteboard there, everything

0:11:47.240000 --> 0:11:48.980000
 is first in, first out.

0:11:48.980000 --> 0:11:52.260000
 If you don't turn on QoS, if you don't
 have any congestion management,

0:11:52.260000 --> 0:11:56.860000
 then you have no control over what's
 going to be dropped if something

0:11:56.860000 --> 0:11:58.420000
 needs to be dropped.

0:11:58.420000 --> 0:12:01.660000
 You have no control over what's going
 to be delayed and held back in a

0:12:01.660000 --> 0:12:05.760000
 queue. If something
 needs to be delayed.

0:12:05.760000 --> 0:12:11.540000
 So, first in, first out is a bad thing
 when we're talking about congestion,

0:12:11.540000 --> 0:12:15.340000
 because that means no control.

0:12:15.340000 --> 0:12:19.420000
 Incoming bursts can cause
 congestion of queues.

0:12:19.420000 --> 0:12:27.020000
 All right, and just to finish this up
 here, I'm not going to go into the

0:12:27.020000 --> 0:12:30.140000
 differences of these, but I just want
 to give you a real quick look here

0:12:30.140000 --> 0:12:35.540000
 at some terms. So, if you ever see these
 terms right here, so FIFO, that

0:12:35.540000 --> 0:12:40.240000
 stands for FIFO, or first in, first
 out, we now know that's bad.

0:12:40.240000 --> 0:12:46.800000
 If you see weighted fair queuing or
 WFQ, low latency queuing, or class

0:12:46.800000 --> 0:12:49.400000
 based weighted fair queuing.

0:12:49.400000 --> 0:12:51.240000
 These are all methods of queuing.

0:12:51.240000 --> 0:12:55.060000
 These are all features that either let
 you determine how many queues are

0:12:55.060000 --> 0:12:57.540000
 going to be available per port.

0:12:57.540000 --> 0:13:03.500000
 And, or let you determine what data
 will be allowed into each of those

0:13:03.500000 --> 0:13:09.220000
 queues. As far as scheduling features
 are concerned, there's round robin.

0:13:09.220000 --> 0:13:12.340000
 Round robin just simply means, okay,
 I'm going to pull one packet from

0:13:12.340000 --> 0:13:14.560000
 each queue, and then I'm
 going to move back again.

0:13:14.560000 --> 0:13:16.980000
 So, all the queues are basically
 treated equally.

0:13:16.980000 --> 0:13:20.620000
 So, there's really no QOS when we're
 talking about round robin.

0:13:20.620000 --> 0:13:23.380000
 There is weighted round robin.

0:13:23.380000 --> 0:13:26.680000
 I say, okay, each queue is given
 a certain weight value.

0:13:26.680000 --> 0:13:30.600000
 The higher the weight, the longer I
 stay on that queue pulling traffic

0:13:30.600000 --> 0:13:35.500000
 from it. There is low
 latency queuing.

0:13:35.500000 --> 0:13:38.940000
 Low latency queuing means typically
 that I've got one queue that's like

0:13:38.940000 --> 0:13:40.680000
 a strict priority queue.

0:13:40.680000 --> 0:13:45.020000
 I'm going to empty that queue out first
 before I look at anybody else.

0:13:45.020000 --> 0:13:47.000000
 So, as long as there's something in
 the low priority queue, I'm going

0:13:47.000000 --> 0:13:50.900000
 to empty it out, and then I'll move
 on to the other queues and deal with

0:13:50.900000 --> 0:13:56.160000
 them. And then there's class
-based weighted fear queuing.

0:13:56.160000 --> 0:13:59.840000
 So, notice that both low latency queuing
 and class-based weight fear queuing

0:13:59.840000 --> 0:14:01.960000
 fall into both categories.

0:14:01.960000 --> 0:14:06.040000
 Those are both, they both have elements
 of queuing from a QOS perspective

0:14:06.040000 --> 0:14:10.020000
 as well as scheduling from
 a QOS perspective.

0:14:10.020000 --> 0:14:14.640000
 So, that concludes this video on congestion
 management with particular

0:14:14.640000 --> 0:14:17.960000
 emphasis on queuing
 and scheduling.

0:14:17.960000 --> 0:14:18.700000
 Thank you for watching.