Cisco Networkers 1998

home *** CD-ROM | disk | FTP | other *** search

/ Cisco Networkers 1998 / Cisco Networkers 1998.iso / pc / talks / nwk05a / nwk05a.eml < prev next >

Wrap

Text File | 1998-04-08 | 59.2 KB | 943 lines

<HTML> <HEAD> </HEAD> <BODY> Hi, my name is Susanne East. I'm a Global Consulting Engineer out of the New York City office for Cisco Systems, and I'm here today to talk to you about ATM, or Asynchronous Transfer Mode. This is not the ATM, Automatic Teller Machine, that you might go to. Unlike the automatic teller machine, which you take money out of, this is something that you put money into. Ha, ha, ha. So, anyhow, as I said, I'm a global consulting engineer out of New York City, and here's my e-mail ID if you need to get in touch with me at any point, or if you have any questions. This agenda here is pretty full, so, what I'd like you to get out of the course, though, is just kind of a - to be kind of buzzword compliance, if you will - compliant, if you will. When you leave the course, I just want you to have an idea of, if you hear something like, SSCOP, you'll know what the acronym means and where it fits in into the whole ATM story. So, we'll start out with a quick introduction, and then we'll speak in length about ATM fundamentals. And then we'll talk about some ATM transport standards, things like UNI and NNI, and so forth. And then we'll talk about campus ATM internetworking, because, actually, this course will focus mainly on ATM as it pertains to the campus. So, we will be talking about things such as LANE, or LAN Emulation, and MPOA, or Multi-Protocol Over ATM - not in great depth, but we will touch on those topics. So, when we talk about ATM, or Asynchronous Transfer Mode, what we're talking about is a technology which uses the small fixed-sized cells of 53 bytes each, in order to transport data from point A to point B. And, it can actually be voice, video, or data, because - the reason being is that, because we're using these fixed-sized cells, we are able to see low latency with predictable throughput. So, this is very - this helps with voice, video, and data all traveling over the same network. And, as we know, this ATM is applicable to both the LAN and the WAN environment. Now, before we jump into all of the ATM details, what I'd like to do is, first, talk about why it came about and why we are seeing such a - such an influx of ATM in our networks today. Basically, here we have a traditional network. And what we have here are, basically, shared Ethernet clients off of a router, and this might be - this is a very hierarchical design, and the broadcast control is done in a very logical manner. We're also using, in this design, an 80/20 rule, is what it's called. Eighty percent of the traffic is local and 20% goes off the network in order to gain - to get to network resources. And, another characteristic of this traditional network is that the user communities are homogenous. And, what is meant by that is that it's really this one-size-fits-all bandwidth. So, it doesn't matter if you're a trader trying to make a million dollar trade, or if you're just some, you know, user who's trying to send an e-mail to his friend. You're getting the same bandwidth and you're contending for the same bandwidth. Additionally, moves, adds, and changes in this scenario are done manually, and you're kind of constrained to the physical constraints of the environment, rather than just being able to do logical moves, adds, and changes. So, how does this diagram evolve? Well, basically, the network hierarchy is still maintained, but, with the quantum leap in performance that switching has brought us, we can now do switching to the desktop. So, we need to keep making the pipe, as it goes through the network, fatter and fatter, in order to accommodate the switching to the desktop. One of the applications that has brought this huge change in networking is that thing called the "World Wide Web." I'm sure everyone has heard of it. If you haven't, you need to turn this off and go read a book on that first. But, at any rate, what we're seeing is this migration in traffic patterns because of the World Wide Web, whereby now 80% of the traffic is going off of the local network and only 20% is remaining on the local network. So, these centralized high-performance servers are now in place for, you know, Web serving, e-mail, and so forth. So, the other quick thing to mention here is that moves, adds, and changes are also being automatted - automated, rather - through the advent of VLANs, or Virtual LANs, or ELANs, Emulated LANs, as they are referred to in ATM. These are a list of acronyms that I won't go through, but, basically, the next few slides are - should just be there for your reference, so you can refer back to these if you need to - if you forget an acronym or if you need to just refresh yourself. So, these slides are just for reference, we won't go through them one by one. At any rate, let's continue on now and talk about some rudimentary ATM concepts. With rudimentary ATM concepts, I'd like to cover four areas: the physical layer of ATM, the signaling layer of ATM, the cell format, and connection types. So, starting out with the physical layer, what I'd first like to do is compare and contrast two concepts that you hear quite often in ATM. One is known as SDH, or Synchronous Digital Hierarchy, and the other is known as SONET, or Synchronous Optical Network. SDH was developed by the CCITT, or the Consultative Committee for International Telephony and Telegraph, and SONET was developed by Bell Labs in the U.S. Incidentally, CCITT was located in Europe at the time. SDH and SONET are two different ways to transport ATM, if you will. And, SDH uses a framing mechanism known as STM, or Synchronous Transport Module. And S - and SONET - uses STS, or Synchronous Transport Signal, and OC, or Optical Carrier. And, what you can see here by this slide, is that this is really just a big multiplexing concept. Where an OC1, for example, would be 51.84 megabits per second, and an OC-3 would be three times that, or 155.52 megabits per second. So, at any rate, what we're going to do here is, mainly in this talk we'll be referring to OC, so OC-1, OC-3 - mainly OC-3 and OC-12, actually. Incidentally, I want to mention two more things, the first being that now the SDH and the SONET standards both reside under a committee, a new committee, relatively new committee, known as the ITU, or the International Telecommunications Union. So, that is all under one umbrella now for better interoperability and so forth. Secondly, a frequently asked question about this slide is, sometimes you see notations that looks like OC-3 and then a little "c" next to it. And what that stands for is "concatenated." So, what that means is, that that OC-3, rather than taken as three 51-megabit-per-second pipes, is taken as three times 51 meg, or 155 megabit per second. So, that's all that the - that's what the "c" stands for, and that's a commonly asked question in this talk. So, moving along with the physical layer. I like to make an analogy with ATM with my car. I just bought - actually last year - I bought a Toyota Rav4 and I absolutely love it. I drove from San Francisco to New York City with it and it's been great. But one thing that I noticed is, because it's four-wheel drive, I can actually drive it up in the mountains and I can drive it, you know, to the beach, and I can drive it just about anywhere. And, oh, and ATM is much the same way. And what you can see from this chart is that ATM is very versatile, in that you can run it over multi-mode fiber, single-mode fiber, coax, UTP, STP, etc. And, what I've listed here is some of the rates and the standards by which these are transported. And - but this is one of the big attractions to ATM. And this is why network managers look to ATM to solve their problems, because they don't have to change anything in the physical infrastructure. So, that was the physical layer, just a brief overview of the physical layer within ATM. Let's move on now and talk about signaling. And this is one of the most important concepts to grasp within ATM. The first thing we'll talk about with ATM signaling is the UNI versus the NNI, the User-to-Network Interface, the UNI, or the Network-to-Network Interface, or the NNI. And, then we'll talk about the different variations of virtual connections and virtual paths. First, touching upon ATM signaling, what we have here is - where you see arrows pointing from - see the router there pointing into the ATM cloud, and the workstation there pointing into the ATM cloud, that's known as a UNI connection, or a user-to-network interface connection. So, any non-ATM switch connecting into an ATM switch would be known as a UNI connection. And, any connection that is between two ATM switches would be called an NNI connection, or network-to-network interface connection. In addition, you might hear a couple of other terms scrambled in here. One is known as the public UNI, which is the public user-to-network interface, and I've notated it on this chart here. And you can see that what that does is, it interconnects the private ATM network with the public ATM network. And, in addition, if you have two public ATM networks, the connection between those is called B-ICI, or Broadband Inter-Carrier Interface - Broadband ISDN Inter-Carrier Interface. At any rate, we'll be focusing mainly on the UNI and the NNI. And, when we talk about UNI later in this talk, I'll cover UNI 3.0, and 3.1, and 4.0, which were all standards that were developed by the ATM Forum. And we'll also talk to NNI more specifically as it pertains to PNNI, and IPNNI, and so forth, and that's the Public NNI, basically. And, one more thing to note here on this slide, I just want to say it out loud, is that the cell header content will vary - and we'll go over this in detail - depending on whether or not it's a UNI cell or a NNI cell. The next major topic within ATM that's important to understand is this concept of virtual paths and virtual channels. In this slide, you can see that the virtual channels are out toward the end, they kind of look like fingers, maybe, out at the end there. And, a virtual channel is basically a logical path between two ATM end stations. And, a virtual path is a grouping of those virtual channels. And then a virtual channel connection is actually a grouping of those virtual paths. So, you see this very hierarchical approach that's involved with ATM. Incidentally, this virtual path and virtual channel becomes very important, and they are actually identified by what are called, "Connection Identifiers," or VPI/VCI numbers - you may have heard to them referred. VPI being Virtual Path Identifier, and VCI being the Virtual Channel Identifier. Let's take this concept one step further to understand where these VPI and VCIs fit in. The main job of the ATM cell switch, which is in the center of the diagram there, is to swap out these VPI/VCI values. That is the main job of the ATM cell switch. What we see - where we have this forwarding information base, where we have the input, and the output, and the port, and the VPI/VCI number, and all those little numbers in there. Basically what the job of the cell switch is this. The cell switch says, "If I see a cell coming in as VPI/VCI 64 on port one, I'm immediately going to my forwarding information database, and I will see that if I get VPI/VCI 64 coming in on port one, I need to switch it out port three as VPI/VCI 29." And, it makes that lookup as fast as possible with very, very little latency. And, essentially, the other thing that should be noted here is ATM is very scalable, because these VPI/VCI values are locally significant down to the port level. So, in other words, what you see here as VPI/VCI 29 on port one, has nothing to do with VPI/VCI 29 that's on port three, because they're - they are each significant and each number means something within itself on that particular port. So, that's a very important concept to grasp as well. Now, being from New York City, when I look at this slide I think of a subway map. And I think it's a good analogy because it really gives you the ability to look at VP switching and VC switching, or Virtual Path switching and Virtual Channel switching. So, let's talk about that for a minute. If I have VCI one on port one and it's, right now, defined to VPI four, the concept of VP switching is that, as that virtual channel one moves through the network, it will still be virtual channel one when it reaches port three, its destination. And, the only thing that will change in this scenario is the VPI number. So, this connection would start out as VPI/VCI 4.1, and when it got to its destination, port three, it would be VPI/VCI 5.1. So, in other words, there's a lot less overhead involved here because we're only having to swap the VPI value. And, at the same token, if we look at the VC switch concept. If we look at VCI one on port one that is on VPI one, this would be analogous to a user - to a person on the subway having to get off the train and switch trains and get on a different seat and so forth, because what we have here is the VPI/VCI value has to change through the VC switch. So, if we have a VPI/VCI one on port one and it's destined for port three, it has to change values on the VC switch to VPI/VCI three and then go out port three as VPI/VCI three, 3.3, actually. So, this concept of VP switching versus VC switching. And this will become very important when we talk about the cell header, but for right now let's look at this in what would look like a real network scenario, if you will. So, in essence here, what we see is that VC switching applies at the UNI. So, we have a cell coming in at the UNI, and it's coming in as VPI/VCI 1.1. As it passes through that UNI switch, which is now a VC switch, we know both the VPI and the VCI values have to change, and here they change to 2.44. But, then, through the NNI switch in the middle, which is notated as the VP switch, the VPI/VCI values become 26.44. So, that VCI value remains constant. So, it's this logical grouping of virtual channels within the virtual path that makes ATM a desirable protocol as well. Now that we've talked about the physical layer, and we've talked about signaling, and we've talked about UNI and NNI, and so forth, we - I'd like to move on now and talk about the cell format and how it pertains to what we've talked to so far. Essentially, when you - when we refer to ATM cells - a term that you often hear regarding that is SAR, or Segmentation and Reassembly. And, essentially, you might hear someone say, "How fast is your SAR chip?" Because if you can't chop up the cells - if you can't chop up the packets into cells fast enough, then if you - it doesn't matter if you have an OC-192 pipe, you know. So, segmentation and reassembly is a very important concept to grasp. So, an ATM cell is basically made up of 53 bytes, and speaking very simply the 5-byte - there's a 5-byte header and 48 bytes of payload. And, incidentally, there was a little argument at first, when the ATM Forum was erected, to figure out whether or not this, you know, how long should this cell be. And, incidentally, the U.S. wanted it to be 64 bytes because it would be a power of two and divisible by four, and the Europeans wanted it to be 32 bytes because it fit more in line with their telephony hardware at that time. So, they settled on 48 bytes of payload and decided, you know, 10% of that could be the header, so we ended up with 53 bytes, which was kind of evenly bad for everybody to reconstruct their hardware and what not. But at any rate, 53 bytes is what a cell is made up of. Now, something that I mentioned before - now that we're going into cell header details, something that I mentioned before is that the cell header will vary depending on whether or not it is a UNI cell, or a user-to-network interface cell, and an NNI cell, or a network-to-network interface cell. So, what we see here is a - the ATM UNI cell has a generic flow control field which is about - which is four bits long, and then we have the VPI field which is eight bits long. So, this is where the NNI and the UNI cells really differ from each other, and what we see is that the VPI value in the UNI cell is much less than that of the NNI cell. So, you can get 256 combinations of VPI for the UNI and 4096 combinations of VPIs for the NNI. And, that's important when we go back to that concept of VP switch versus VC switch, and where the VP switch would be available at the NNI, for example. So, that's where that comes in. That's why that is significant. Incidentally, that generic flow control field in the UNI cell that you see there is not actually used. They thought that it would be used for congestion control and flow control, but that just never came to fruition. So that piece - those four bits aren't not - are not yet defined per se. The remaining part of the ATM cell header for both the UNI and the NNI cell are exactly the same. They each have a 16-bit VCI field. And, then they have this field called the PTI, or the Payload Type Identifier field. And, that's actually comprised of three bits. The first bit says, "Is this user data or is it control data?" And, then the second bit says, "If it's turned on, it will say that there is congestion in the network." So, that's a congestion-indicator bit and we'll talk more about that as it pertains to traffic management later in this talk. And, the third bit in that field is known as the last cell bit. And that you'll see mainly in, what we call, AAL5, or ATM Adaptation Layer Five, which we'll also go into at a later time, but I just wanted to get that buzzword in your head. So, when we talk about AAL5, you'll think of that PTI field and the third bit. Finally we have the CLP bit, or the Cell Loss Priority bit, similar to the disk card eligibility bit in frame relay. If that bit is turned on, it is eligible to be discarded, so that is what the clip bit and - or clipper bit, as they call it - in ATM does. And, then finally we have an 8 bit CRC header error check field. So, now we've talked about virtual paths, and virtual channels, and we've talked about the cell header, and now I'd like to talk about the different types of connections that are available within ATM. I mentioned, in the very beginning, that ATM is connection oriented, and now we're going to go into more detail on how those connections are actually derived. We have three different types of connections that we'll be covering: the permanent virtual circuit, the switched virtual circuit, and the soft permanent virtual circuit. Before we do that though, I'd like to just make a quick distinction between connectionless packet routing and connection-oriented cell switching, because, although they are vastly different, it is good to compare them. Because most of us are more aware of connectionless packet routing. So, if we compare them, it becomes pretty obvious, the differences. So, in a regular routed network that you might have worked on in the past, you might have worked on a network whereby you had Cisco routers running, you know, RIP or, you know, with TCP/IP and what not. And, basically what was - what the case is there is that the data can take different paths to get from source to destination. The burden is on the packet, if you will, to get from source to destination and it's this hop-by-hop routing. And, in addition, each packet can take a different path every time, and there is also a burden placed on the destination in that the destination has to reorder all those packets and put them back together so that they make one cohesive data stream, if you will. Let's contrast that with connection-oriented cell switching. In a connection-oriented cell switching environment, what happens is source one will actually set up a call to source eight, if you will, or destination eight, as it's listed here. And, the data must be - the connection is made - the data must take the same path and arrive in sequence when it gets to the end station. So, there's a lot less burden on the destination to be putting cells back together and what not with regard to ordering. So, once that connection has been made and the data's been transferred, in some cases that connection is then brought down, so then the provision bandwidth can be used elsewhere and what not, and we'll talk about that in a moment. So, packet routing versus cell switching. Cell switching must take the same path each time and the cells must arrive in order when they do reach the destination. So, first, we'll talk very briefly about the Permanent Virtual Circuit and - or PVC, sometimes known as a Permanent Virtual Channel, or a Permanent Virtual Connection, but PVC - Permanent Virtual Circuit. And, what happens here is this. Previously, I talked about that the job of the ATM switch is to look in its forwarding information database and figure out which VPI/VCI values to swap. Well, in this permanent virtual circuit scenario, what would have to happen here is that if - for each hop in the network - the network administrator would have to set up by hand the per - the forwarding-information database. So, if I had workstation A talking to workstation C, what he - what will have to happen here is the network administration - administrator will have to go to the first switch in the path and actually type in, "If VPI/VCI 29 comes into port 1, switch it out port three as VPI/VCI 45." And, then they have to go into the next switch along the path and say, "If VPI/VCI 45 comes into this switch on port one, switch it out port two as VPI/VCI 16," and so on until the actu - until the packet actual - or the cells actually reach the destination, which is C in this case. So, the connection is always nailed up, if you will, and the - and it's high overhead you can see involved here. This is still used. It's a little archaic for ATM technology, but it is still used, mainly to interconnect public - private ATM networks with the public ATM infrastructure, if you will. So, that's a permanent virtual circuit. And, now we have a little - we take this a little bit to one more sophistication level, if you will, and we have a switched virtual circuit. And, what a switched virtual circuit does is, it actually uses that UNI and NNI signaling protocols in order to build the connection from end to end. So, when workstation A wants to wor - talk to workstation C, in this scenario, A will actually speak to the switch that's the UNI connection via UNI signaling and say, "Hey, I want to establish a connection with workstation C." And, then the route, if you will, is built automatically and dynamically through the network using the NNI signaling. And then the UNNI - or the UNI - from the, to the destination workstation and the - through that signaling the connection is then built, as we can see here. And the forwarding-information database tables are filled in automatically by the protocols. And then, once that conversation is over, there's kind of a hang-up that happens and that connection is then torn down, as is shown here, and the database is cleared out and now you can use that provision bandwidth for something else. So, this concept of switched virtual circuits is very relevant in most of campus ATM networking, such as LAN emulation, for example. So, that is the switched virtual circuit. Next, we have this hybrid solution, if you will, known as a Soft Permanent Virtual Circuit, or an SPVC, if you will. And, what happens with an SPVC is this. What you can see here is that the UNI - the connection is built statically, or a static route, if you will is - a static ATM route is built. But then you let the NNI signaling in the center of the ATM cloud build the connection dynamically through the ATM NNI cloud, and then the connection is manually built at the UNI at the other end. So, what this does, is it - it's kind of the best of both worlds. So, if you have - what you'll notice now is station A and station C have become routers. And that's significant because these soft PVCs are mainly used for devices such as routers, or servers which have a low tolerance for the call-setup latency that they might experience. Especially if they know that all packets destined, you know, out - it's kind of like a route of last resort, for example. Then it's easier just to set up a soft PVC and, therefore, the router doesn't have to do any overhead with regard to UNI processing and so forth. So, that's known as a soft PVC. So, the PVC is established manually across the UNI at both the source and the destination. And then the ATM cloud in the middle, or the NNI cloud, that's established dynamically. So, you can reroute and so forth in that area. Next, what I'd like to do is kind of cover what we've looked at so far as it pertains to the ATM reference model. So, a little bit of review and then we'll take it from there, when we talk about the ATM adaptation layers, you'll see. So, the ATM reference model is actually a three-dimensional model, but I've boiled it down into basically three layers. There is the physical layer of ATM, and then there's the ATM layer of ATM, and then there's a new thing called the ATM adaptation layer, well, not new, but new to the audience maybe. ATM adaptation layer. So, when we talk about the physical layer, what you may come across in your reading or your research is that there are two sublayers. There is a Transmission Convergence sublayer, or the TC, and there's a Physical Media Dependent sublayer, or the PMD. Basically the PMD, or the physical media dependent sublayer, is in charge of media coding and putting the bits on the wire. And, also, it is in charge of telling the transmission convergence sublayer which - what kind of framing should be used when the transmission convergence sublayer actually frames the cells into whether it's a copper medium, or a fiber medium, and so forth. So, the TC and the PMD. And, this chart should look familiar, and when you think of the physical layer of ATM, you can think of this chart and all of the different data rates and the media type and the framing that's involved within ATM. So, this is just an eye chart for reference. Moving along through our review. When we talk about the ATM layer, we've actually already talked about some of the different things that happen at the ATM layer, the first being the cell header insertion and removal. And, with the cell header insertion and removal, what we're really talking about is swapping out that VPI/VCI value. So, when you think of the ATM layer, you can think of that. Now that swapping out of the VPI/VCI value along switches in a path is also known as cell relay. I think I neglected to mention that before, but cell relay is that swapping of VPI/VCI value. And, then finally, the ATM layer will also multiplex and demultiplex cells of different connections through a path. So, here's another slide that should look somewhat familiar. So, this - in this scenario here - the ATM layer would provide the VPI/VCI values in the header, and it would also ensure that the cells stay in the proper order. And, anything that has to do with the cell header, like the cell loss priority bit, for example, would be handled at the ATM layer as well. Finally, a new concept here, the concept of ATM adaptation layer. And, within the ATM adaptation layer, there are actually two sublayers. There's the Convergence Sublayer, or the CS, and something that we've already talked about - the Segmentation and Reassembly sublayer, or the SAR. And, when we talk about these, the convergence sublayer and the SAR, essentially when we talk about the SAR functionality, we've already talked about the fact that the SAR will actually take a packet and chop it up into those 48 byte pieces of payload, or PDU, it's actually called, primary data unit. And, basically, what the convergence sublayer does is it tells the SAR how to chop up those packets. So, in other words, it adds what's called a cell tax. So, it's not so cut and dry. That 48 bytes of payload could either be 48 bytes of payload, or, depending on the cell tax, some bytes could be stolen from that payload field in order to accommodate different types of traffic. Now, what am I talking about here? You've probably heard about all the different types of AALs that there are available. There's AAL1, AAL2, and then AAL3/4 were combined, so now they're AAL3/4, and then there's AAL5. So, when we talk about AAL1, we're talking about this process whereby, when the cell is actually chopped up into - when the packet is actually chopped up into cells - one byte per cell is set aside for things like timing and synchronization bits for what we call CBR, Constant Bit Rate, applications mainly. So, AAL1, when we talk about constant bit rate in a moment when we get to service categories, would apply to constant bit rate. Secondly, we have AAL2, and AAL2 actually was never completely ratified by the ATM Forum. Where we're seeing AAL2 rear its head these days is with regard to something called "composite unit." And, what's actually happening with this composite unit concept is that, in the VTOA, or the Voice and Telephony Over ATM forum, if you will, is actually coming up with this idea whereby cells, which were once 48 bytes of payload, might now be, you know, less than that. And, basically, they haven't come to a decision yet. But making these cells smaller than 53 bytes, essentially, is their goal, such that they'll be able to accommodate voice and video in a better fashion. So, AAL2 is really not big. That's what I would say. It's not really done yet. It's - it may never get done. And then, moving along, we'll look at AAL3/4. And AAL3/4 is very high overhead, as you can see. Essentially, AAL3/4 is for what we call SMDS traffic. And, essentially, what happens here is that, when the SAR chops up the packet into cells in AAL3/4, it actually reserves 2 bytes of header and 2 bytes of a trailer that are taken from the 48 bytes of payload. So, now you only have 44 bytes of payload to work with. And, so, suchly, this isn't a very good way to network because you're using most of your payload field where you're going to actually put real information for overhead bits, if you will. So, incidentally, with this whole story, AAL5 happens to be the clear winner in the AAL story. And, AAL5 actually has no cell tax. When the packets are chopped up in the cells, there's no cell tax applied. But rather, what happens here is the last cell in a stream of cells is set aside to be able to do a 32-bit CRC check against all the other cells when they're put back together at the destination. So, rather than taking from each cell along the way, it just has one big cell at the end that does the check for all the other ones ahead of it. So, remember when I mentioned that PTI field, the payload type identifier field, that last bit, the last cell bit? When that's turned on, we know that that's AAL5. So, this concept of cell tax is very important. AAL5 is also known as Native ATM, or sometimes you might hear it, or see it, or read it referred to as SEAL or S-E-A-L. Simple and Efficient Adaptation Layer is what it was called originally. So, those are the AALs. And, I'll refer back to them when we talk about the different service categories and we talk about constant bit rate, and variable bit rate, and so forth. So, that was quite a bit of information to digest, but now what I'd like to do is take what we've learned so far and talk about a day in the life of a cell with regard to ATM. So, in this little diagram here I think this helps put it together, like where ATM fits in in the whole big picture. And, essentially, we have a regular TCP packet with the TCP header and the application data. It then moves down the stack to the IP layer, where it becomes an IP datagram by fixing the AT - the IP header. And, then, at the data link layer, we add LLC/SNAP information for protocol identifiers and so forth. And, then we get into the ATM piece of it where the actual convergence sublayer is applied, cell tax is applied, and the segmentation and reassembly happens at the AAL layer. And, then the ATM layer is where the - now we have all these little 48 bytes of payload or PDUs, if you will, primary data units, coming into the ATM layer. And, at that ATM layer, that's where five - the 5-byte header is added, the - excuse me - the VPI/VCI value, the CLP, or cell loss priority bit, is either turned on and so forth. So, that's at the ATM layer. And, as we move down the stack, we get into the physical layer where the transmission convergence sublayer and the physical media dependent sublayer take over, and the bits are put on to the wire. So, in a nutshell, this is a very easy way to think about how ATM fits into the whole picture. And, incidentally, there's a lot of debate on whether or not it's a Layer 2 protocol or a Layer 3 protocol and so forth. I think of it as a Layer 2 protocol myself. But it does have some of the complexities, as you'll see when we talk about ATM routing and Quality of Service of, you know, it's - and, at times, even more complex than a Layer 3 protocol. So, that debate remains, and it's, you know, not really a religious argument, but... At any rate, let's move on now and talk about those ATM service categories that I mentioned before. When we talk about these service categories, it's, first, important to define these traffic descriptors and the Quality of Service parameters. And then we'll move in and we'll talk about the actual service categories - the BRs: the CBR, VBR, UBR, AVR. I don't think I could say that faster. Now, when we talk about ATM service criteria, what we're talking about are end stations making a contract with the network so that their application can be accommodated throughout the ATM cloud. And, when they make this contract, they actually use different traffic descriptors and different Quality of Service descriptors, if you will. The traffic descriptors listed here - which we will go into - are: Peak Cell Rate, Sustainable Cell Rate, Maximum Burst Size, and Minimum Cell Rate, and the Quality of Service descriptors are delay and cell loss. So, cell delay and cell loss. Let's just touch on these very briefly, because it's important to understand what each of these acronyms means and what not, before we just jump in and talk about the different service categories. Incidentally, these traffic descriptors are defined in something that I'll talk about in a moment called, "Signaling 4.0 Specification." And, they're known as Traffic Management 4.0, or TM 4.0, you may hear it referred to. So, within TM 4.0 or Traffic Management 4.0 specification, there are several definitions for how these different traffic descriptors can be used. So, we'll start with the peak cell rate. And, peak cell rate would be like the application saying to the network, "I need the most amount of bandwidth you can give me for this application." Peak cell rate is usually an indication that it's a CBR connection or a constant bit rate application. So, that's peak cell rate, the maximum data rate at which the - that the application will need in order to accommodate the given application. Next we have the sustainable cell rate, or the SCR. And, what this is - it's like the application saying to the network, "I only need this as an average amount of data transfer bandwidth that I'll need through the networks." So, it's the average amount of bandwidth that's needed by the application. And, then, usually in tandem with sustainable cell rate, is this concept of maximum burst size. So, even though I need this average amount of bandwidth, how many times will I need to burst up to what might be the peak cell rate? So, SCR and maximum burst size, or MBS, are usually used in tandem. And, we'll see that in a moment as well. And, then, finally, we have the minimum cell rate traffic descriptor, or MCR. And, basically, MCR is kind of like the "slacker" of the traffic descriptors. In that, this is an application saying, "Ah, you know, it'll get there when it gets there. Latency is not really an issue. I just need this minimum amount to get my application through." Next, we have the Quality of Service parameters, starting out with the Maximum Cell Transfer Delay, or the MCTD. And, essentially, the maximum cell transfer delay is the amount of accumulated delay from end points in the network. So, that accumulated latency and delay that's seen in the network, that would be defined as the maximum cell transfer delay. And, then, finally, we have the Cell Delay Variation Tolerance, the CDVT. And, this is the ability for a - for an application to be able to handle things like jitter, or, you know, incoming cells coming in at a - at an inconsistent rate, if you will. And, so when, you know, the picture - hopefully this one's not doing it - but when a picture gets, you know, kind of chopped up and the voice quality gets low and that kind of thing, usually the - they're - the application probably is suffering from a cell delay variation problem. And, finally, with regard to ATM Quality of Service, there is Cell Loss. And, there's this concept of a cell loss ratio. And, this is where - you know, there are some applications that can tolerate some percentage of cell loss. For example, now this - right here is where we get into this - the difference really between voice and video traffic versus data traffic. And, when we talk about voice and video traffic, delay becomes an issue with them. For example, if there was a 200 millisecond delay for voice or video, you would see a huge impact in the way that that type of traffic was delivered to the end station. However, if there's a delay in data of 200 milliseconds, it doesn't really make that - it doesn't make a big difference. Cell loss, however, for voice and video, if you lose a couple cells here and there, it's not going to change the overall scope of what you're seeing or hearing. Whereas cell loss ratio with regard to data applications - if one cell is lost from a packet, then that whole - from a string of cells that make up a packet - then that whole packet is useless, and we're going to go into that. So, really, this distinction of cell loss ratio is more applicable to data or it's going to have a lesser - data will have a lesser tolerance for cell loss. Whereas voice and video will have a lesser tolerance for cell delay, if you will. So, now that we've defined all the different traffic parameters and Quality of Service parameters, I'd like to step now into the service categories known as CBR, or constant bit rate; VBR, variable bit rate; UBR, unspecified bit rate; and, finally, ABR, which is available bit rate. So, now we'll apply a lot of what we have learned up to this point to this piece of the talk. With - and this is really the crux of ATM right here, this ability to be able to apply Quality of Service to a connection, for example. So, this is really the true differentiator. This and traffic shaping are really the differentiators between ATM and, like, a technology such as Ethernet. So, with constant bit rate, having said that, constant bit rate is a traffic identifier - a service category, rather - that is the creme de la creme, if you will, of traffic categories - service categories. It's kind of like flying first class on, you know, Singapore Air versus, you know, Tower Air to, you know, from New York to Florida, if you will. But at any rate, constant bit rate applications are there to accommodate such things as realtime voice and video applications. And, the traffic descriptor that's used in order to negotiate with the network is that peak cell rate traffic descriptor. And, in addition, the cell delay and the cell loss tolerance are very low. So, when negotiating with the network for cell loss ratio and cell delay parameters, those will register very low tolerance for that. So, CBR is like, as I said before, like first-class service category. Next, we have a concept known as variable bit rate. And, variable bit rate comes in two separate flavors. Variable bit rate comes in realtime, so VBR-RT, and variable bit rate non-realtime, or NRT. Variable bit rate is used for types of applications that can handle some cell loss and some cell delay, and so forth. Things like packetized voice and video, or SNA, for example. Those are two examples of using VBR. Essentially, VBR will use many more traffic descriptors than CBR. So, for example, it uses actually peak cell rate, and in addition, it uses that sustainable cell rate, which is the average bandwidth that's going to be needed. And, then, finally, the - it uses that maximum burst size, which works in tandem with the SCR. And, then what we see on our little Quality of Service tolerance chart here, is that cell loss - I'm sorry, cell delay will vary depending on whether or not it is a realtime applicati - realtime VBR or non-realtime VBR. So, for example, cell delay for a realtime application is going to be very low. Cell delay tolerance for a non-realtime application is a bit higher. And, cell loss for both of them is pretty much on a lower part of the chart, because it is mainly for different types of applications that do need some semblance of Quality of Service. So, moving along, next we'll talk about UBR, or unspecified bit rate. An unspecified bit rate is really the - pretty much - mainly what's used, kind of like how the Internet is today. Where, you know, you send an e-mail. You're not sure it's going to get from source to destination, but you can be pretty sure that it's going to get to its destination. So, the way that I describe it is, that it's kind of "send and pray." Incidentally, I read a funny quote the other day about the Internet. And, it said that the Internet works in practice, but not in theory. And, I thought that was kind of funny and very applicable. So, at any rate, moving along here, UBR, which is kind of, you know, probably the lowest category that you can go, as far as these service categories go, for Quality of Service, has very high tolerance. It will register very high tolerance for cell delay and cell loss, for example. And, then finally, probably the most commonly used of these service categories for data is ABR, or Available Bit Rate. And, as it's listed here for the applications, ABR's mainly used for LAN interconnect and for data transfers. It uses peak cell rate and minimum cell rate as its traffic descriptors. Note that minimum cell rate is used because it's data, and it only needs minimum - the cell rate can be minimum because the cells can arrive at any rate, if you will. But, if you notice on our Quality of Service chart, cell loss tolerance is very low. So, in a data network, it's important to have very low cell loss tolerance, but delay can be, you know, it doesn't matter when it gets here as long as it all gets here. That's the concept here. One thing that I'll talk about with regard to available bit rate is this concept of congestion feedback mechanisms. And, we'll get into that in greater detail when we talk about the service categories. Oh, incidentally, going back a frame here - when we talk about CBR, VBR, UBR, and ABR. They're - when we - I just want to highend those AALs real quick. And, basically, when we talk about constant bit rate applications, we're mainly talking about using that AAL1-type frame format - or cell format, in order to accommodate cell - constant bit rate applications. And, for available bit rate con - applications or any data applications, AAL5 is the standard of choice for that. So, when we think of ABR, mainly thinking of AAL5, and when we think of CBR, mainly thinking of AAL1. And, VBR can really be a mixture of those. </BODY> </HTML>