This chapter discusses ATM, Virtual LAN, and Multimedia on Ethernet (including Gigabit Ethernet.)
Note: Today's notes are being posted with Queries that DO NOT HAVE ANSWERS written out next to them. After Thursday's lecture, these notes will be replaced with Queries that DO have the answers written in. For exam study purposes you should print out those notes, rather than these. However, for class preparation, you should print THESE out and develop your own answers to the queries so you'll be ready in class.
JMM
ATM: Asynchronous Transfer Mode. Designed for computer data, voice and multimedia: videoconferencing, video on demand, etc. ATM supports Quality of Service (QOS) - the ability to guarantee a priori (beforehand) that sufficient bandwidth and adequate latency (low enough for the job) can be provided.
Ethernet's variable-sized packets (called "frames") can be several thousand bytes long. ATM's fundamental assumption is that very short fixed-size cells or packets would increase the "liquidity" of data flow - the ability to reroute it as necessary and assemble the fragments at the other end.Frame switches (for Ethernet or Token Ring) are hard to build beyond 100 Mpbs.
There is an analogy to RISC computing architectures. Less cases (e. g. multiple-length CISC instructions) to deal with, means more silicon available to focus on optimizing the treatment of the short simple instructions you do have.
155 Mbps = OC-3
622 Mbps = OC-12
2.5 Gbps =OC-48
Sampling: Consider the problem of sampling audio at 44k samples/second, video at 1.5 m samples/second and bursty computer data in blocks of 10 Mbytes. Small cell size makes this possible.
QOS: ATM looks for a route and commits "seats" on a series of "airline flights" to guarantee that data can be sent at the requested latency & bandwidth, before accepting a transmission. <See Figure 5.2 in the text.>
Query 13.1: How long might a customer have to wait before the desired QoS is available? Study figure 5.2 to figure out the answer.
BASIC FACT: (p. 163) ATM is a connection-oriented approach that uses a circuit switching design to guarantee transmission reliability. LANs such as Ethernet represent the connectionless packet switching approach, on top of which connection-oriented services are constructed. Reliability in that case is bought by adding the acknowledgement mechanism, which of course adds overhead cost. (We haven't studied TCP/IP in detail yet; that's coming in the Internet chapter.)
ATM cell:
|
|
|
||
| Switch Connection: 32 bits | Header Checksum: 8 bits |
|
|
Cell Switching is accomplished by establishing a "virtual circuit connection" (VCC). That is, the two ends of the VCC perceive the connection as though there were a dedicated wire from one end to the other. One "hop" is called a Virtual Connection (VC), and may consist of a permanent VC (PVC) for a leased line, or a switched VC (SVC) for a dynamically established connection.
A bundle of VCs with same start and destination are grouped together as a virtual path (VP). An identifier for a VP or VC is called a VPI/VCI. These look like 1/3 (meaning path 1, circuit 3).
Consider Figure 5.4. A better picture would have shown multiple ATM switches, so that you could see how a route grows across the network. Here's such a picture.
As can be seen, the data source device is attached to input port 8 of ATM1, which in turn is attached via its output 1 to the input 6 of ATM2. From ATM2's output 8 there is a cable to input 4 of ATM3. The routing to establish a virtual circuit from the data source to the data destination involves telling three switches what to hook where.
The VPI/VCI designators within each switch are like "account numbers" which help the switch keep track of what it's doing for whom. The overall virtual circuit has different VPI/VCI designators within each switch.
Query 13.2: Look at figure 5.4. Why does Input Port 7 have several different output ports listed in its Routing Table?
Query 13.3: Also in figure 5.4. Assume this switch is ATM Switch 1 in my diagram above. What switch needs to know the information in "VPI/VCI translation table for output port 5"? Hint: There may be lots of other switches in the network besides ATM1, 2 and 3 that I've shown above.
When an ATM device tries to establish a VCC, a "signaling packet" is sent out. It contains a specification of the necessary QoS. If it passes all through the route to the destination and all intermediate switches are willing to provide the QoS, then a reply propagates back through the net and the Virtual Circuit Connection is established.
A terminology break. We defined "Isochronous" and "Synchronous" back in Lecture 3, but we ought to refine the definition. Voice and circuit emulation are set forth in Table 5.1 as the paradigm examples of isochronous data, so that anchors our concept: some delay is tolerated but the minimum and maximum delay ("jitter") must both be bounded.
Synchronous is where the maximum delay must be bounded, but there is no penalty for early delivery. Thus, synchronous communications could have larger jitter than isochronous. This implies that a receiver for such a signal needs memory sufficient to store up to (maximum delay) worth of signal.
We also should define 'bursty': it means asynchronous, with the additional property that the average data rate varies greatly across time and can include periods of zero data.
Query 13.4: ATM has five flavors of Adaptation Layers, described on Page 169 in Table 1. These layers are responsible for taking application frames of whatever size, and packing them into ATM cells with appropriate routing set up for delivery. Which flavor of AL is most similar in its purposes to TCP? Why?
Query 13.5: There are four classes of service - CBR, VBR, ABR and UBR. Describe a plausible application which might use each class of service.
Query 13.6: Look at Figure 5.9. In TCP/IP (in which IP packets may arrive via the "connection-less" process through different physical channels and out of order), there is stored in each IP packet some information for re-sequencing the data so that TCP can deliver the data in order. However in the ATM layer there is no evidence of any sequencing control field. How is ATM's convergence layer able to guarantee that the application frame will be reconstructed in proper order?
Query 13.7: What is an edge switch?
Query 13.8: What's the relationship between SONET and ATM technology.
Query 13.9: What is IP Switching?
Query 13.10: LANE and who cares?
Query 13.11: What is MPOA and how does it relate to LANE?
What is ATM used for?
Originally it was intended as a high powered LAN. In fact it is often used as a backbone to tie together multiple LANS, and as a layer in the architecture of WANS. This can be done in increments of 1.5 Mbps to 45 Mbps. And the key issue behind it all is ...MULTIMEDIA.
Ethernet Tricks
PACE: Unless we have ATM to the desktop host, we'll be using Ethernet for the "final feet". PACE (Priority Access Control Enabled) switches can be used with 10 and 100mhz Ethernet to establish QoS. PACE strategies include replacement of the random access collision method with managed techniques ("Interactive access") that don't favor the station that last transmitted successfully. This can move Ethernet effective bandwidth utilization from 70% to 98%.
RSVP: Resource Reservation Protocol, an IETF invention, allows end stations to negotiate with routers for a given flow of packets. Sounds like the perfect Session layer technique for use with PACE.
Cells in Frames: CIF is a software technology that is intended to provide explicit flow rate control by pre-packaging the ATM cells into frames (e. g. TCP frames) so that when they get to the ATM switch, there is little work to do in repacking them into ATM cells. It's like shipping eggs by FedEx in such a way that when a carton of eggs gets to the FedEx station, the eggs will pop out of the cartion and be just the right size to fit into the individual little FedEx envelopes - rather than having to be sliced up and reconstituted at the other end.
TCP uses a feedback rate process whereby it continually tests the net on about a 1 second cycle. If a higher rate is possible, the rate of injecting IP packets increases. If not, it decreases. CIF has a much finer granularity and faster means of managing flow rate.
Gigabit Ethernet. Well, it'll be fast....
Back to previous lecture
Forward to next lecture
Back to the Index
Back to the Syllabus