Administrivia. Homework on class webpage If you are having problems following me in class (or doing the homework problems), please buy the textbook

Transcription

1 Administrivia Homework on class webpage If you are having problems following me in class (or doing the homework problems), please buy the textbook Project Discussion class_ gotcha Reading finally on webpage sorry about the delay Will delay discussion of the papers Midterm next Thursday SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 1

2 Switched Networks How do switches allow scalability? Switch vs. Repeater Switching and heterogeneity Packet switching approaches SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 2

3 Virtual Circuit vs. Datagram Virtual Circuit Need to wait a full RTT for connection setup before sending data Connection request contains full address; subsequent data packets contain only circuit identifier A link or switch fails? tough, need to reestablish path Connection setup provides opportunity to reserve resources Datagram: No connection setup can start communication immediately No way to tell if a path exists Packets are routed independently, it is possible to route around failures Every packet must carry full destination information SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 3

4 Source Routing SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 4

5 So what does a switch look like anyways? I/O bus CPU Interface 1 Interface 2 Main memory Interface 3 Can be built from a general-purpose computer Usually specialized high-performance switches towards the core of the Internet (will look at switch design later) SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 5

6 Bridges/LAN Switches Suppose we want to build an ethernet network bigger than 1024 hosts or using more than 3 populated segments in a row Can we use a LAN switch? Also known as bridges A bridge/lan switch that operates on link-layer frames (as opposed to network layer packets) Idea of bridge is to transparently extend the LAN Is this another way of saying repeater? SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 6

7 Example A B C Bridge Port 1 Port 2 X Y Z Example: two ethernet LANs connected by a bridge, compared to two ethernet segments connected by a repeater The bridge appears as an additional host with its own ethernet address on each segment Two communications can be going on on the two different segments without collision What if the bridge simply forwards all incoming frames to all outgoing segments? Can forwarding be done in a better way? Network administrator sets up forwarding table? SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 7

8 Learning Bridges Idea: if a bridge sees a frame sent by node X coming on port y, it learns that the way to X is through port y What if it sees a frame destined to X appear on y? What if it sees a frame for X appear on a port z? What if it sees a frame it does not know? Dynamically constructs a routing table Any problems with this scheme? SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 8

9 Aside Network Caches The routing table is a cache where information is dynamically kept Caches are used to optimize protocol performance Entries in the cache are tagged with a Time to Live (TTL) Periodically, the TTL value on all entries is reduced by 1; an entry expires if its TTL reaches 0 Each time a frame is seen from a destination, its entry TTL is set to some maximum value Cache has a fixed size (to conserve space and make indexing more efficient) What happens when you see a frame from a new node and the table is full? SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 9

10 Problem Loops A B3 B C B5 B2 D B7 K E F B1 G H B6 B4 I J Problem the network can contain loops Learning can go wrong Frames can get stuck indefinitely in the network What can be done? Rely on administrator to statically eliminate loops? SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 10

11 Spanning Tree Algorithm in IEEE IEEE is the standard for LAN/MAN Bridging and Management Idea: bridges disable some ports to eliminate cycles General Algorithm: Bridges elect a tree root bridge Each bridge calculates shortest path to root Bridges on each LAN elect a designated bridge such that It is the closest bridge to the root Break ties using bridge id SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 11

12 Implementation All bridges initially assume they are root They send out configuration messages on all ports Config. message: (bridge id, distance, root) When bridge receives a config message with a better config: It updates its interface information Adds 1 to distance and sends new config. message on all other ports A configuration A is better than B if It identifies a root with a smaller id The root has an equal id, but shorter distance Root and distance are equal but bridge id is smaller Bridge only forwards messages on interfaces if the configuration lists them as designated bridge SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 12

13 Example A B3 B C B5 B2 D B7 K E F B1 G H B6 B4 I J B3 receives (2, 0, 2) from 2; accepts 2 as root (2 < 3) B3 sends (3, 1, 2) to 5 B5 receives (1,0,1) from 1 and sends (5,1,1) to 3 B3 accepts root as 1, realizes B2 and B5 are closer, stops forwarding on both interfaces SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 13

14 Discussion Easy to administer, but optimal? Bridge ids have to be unique What happens if a bridge or link fails? Heartbeat messages/ttl What if two LAN segments are different speed? (e.g., 10Mbit ethernet and 100Mbit ethernet) What if two segments use different Link protocols (e.g., FDDI and Ethernet)? What to do with broadcast and multicast packets? Can bridge learn multicast groups? Transparency (an extended LAN appearing as a regular LAN) is dangerous why? Why not build the Internet out of bridges/switched LANs? SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 14

15 Limitations of Bridges Routing is suboptimal costly for large networks No mechanism to impose hierarchy Broadcast floods the whole network No support for heterogeniety SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 15

16 Discussion Virtual Circuit Switching? Why? Seems counter to the Internet Architecture papers perspective? Phone companies (Telecoms) powerful players, they wanted to play a role in data transport ATM (Asynchronous Transfer Mode) was the results VC (to better support voice) Fixed packet size (cell switching) for simpler faster router design Small cell size (to better support voice and provide more pipelined use of network) Because cell size is small, IP doesnt fit in it Add another layer (Application Adaptation Layer) to provide bigger frame size Heavily used for WANs, but not popular for LANs SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 16

17 Virtual LANs W X VLAN 100 VLAN 100 B1 B2 VLAN 200 VLAN 200 Y Z Logically configure parts of the extended LAN to form a virtual LAN Packets sent on one VLAN segment will be visible to all segments of the VLAN (but nowhere else) When a packet arrives at a bridge port with a host id as sender, the bridge inserts the VLAN id associated with that port Advantage: Enhances scalability why? Disadvantage: Requires administrator intervention SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 17

18 What is Ahead Internet Protocol (IP) the solution to the limitations of bridges Before then ATM important cell switched technology. Not directly compatible with IP Example of VC switching SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 18

19 Asynchronous Transfer Mode (ATM) Virtual Circuit cell-switched technology Packets are fixed size and small Why fixed size? Cell switching more efficient: simpler; allows parallelism Conducive to Quality of Service guarantees What size? If large, low utilization for small messages If small, high header-to-payload ratio (overhead) and more cells to process Why small? Share the link at a finer grain; can preempt for a high-priority packet Store and forward more efficient with small packets; close to cut through Dont have to delay too much to fill a cell with voice samples SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 19

20 Cell Format (48 bytes) GFC VPI VCI Type CLP HEC (CRC-8) Payload User-Network-Interface (UNI) host to switch format GFC: Generic Flow Control, for arbitrating access if shared medium VPI (Virtual Path Identifier) and VCI (Virtual Circuit Identifier), used to identify the virtual connection Type: management, congestion control, AAL5 (later) CLP: Cell Loss Priority: if 1, ok to drop on congestion HEC: Header Error Check (CRC-8) Network-Network-Interface (NNI): used among switches Replace GFC with 4 more bits of VPI SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 20

21 Segmentation and Reassembly AAL AAL ATM ATM For Data networks, upper protocols deal with variable size/larger size packets think sockets Need a segmentation and reassembly (SAR) capability Implemented in the ATM Adaptation Layer (AAL) Different AALs defined for different types of traffic AALs also implement the services exported to the application AAL 1 and 2 (fixed rate); AAL 3/4 (packet oriented); AAL 5 (improvement on AAL 3/4) SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 21

22 AAL 3/ < 64 KB CPI Btag BASize User data Pad 0 Etag Len Convergence Sublayer Protocol Data Unit (CS- PDU) CPI Common Part Indicator (version) BTAG/ETAG beginning and ending tags BAsize: hint on the amount of buffering to allocate (why not actual size?) PAD: user data padded to multiple of 3 bytes Length: size of the full PDU How to translate this PDU to ATM cells? SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 22

23 Conversion to Cells Add AAL Header (44 bytes) 6 10 ATM header Type SEQ MID Payload Length CRC-10 Type used to indicate whether the cell is: BOM: Beginning of Message COM: Continuation of Message EOM: End of Message SSM: Single Segment Message SEQ: 4 bit sequence number to protect against lost/misordered cells (what if 16 consecutive cells are lost?) MID: Message ID is used to multiplex link among multiple messages Length: Length of the full PDU SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 23

24 AAL 3/4 Discussion CS-PDU header User data CS-PDU trailer 44 bytes 44 bytes 44 bytes 44 bytes ATM header AAL header Cell payload AAL trailer Padding What shortcomings can you see in this AAL? SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 24

25 AAL 5 < 64 KB 0 47 bytes Data Pad Reserved Len CRC-32 PAD so that the trailer always falls at the end of ATM cell Length is the size of the Data CRC-32 detects missing or misordered cells Cell Format: Very simple: use a single end of PDU bit in ATM header Type field Full cell payload (48 bytes) used for data What do we lose/gain vs. AAL 3/4? AAL5 has taken over as the de facto standard SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 25

26 AAL5 SAR Padding User data CS-PDU trailer 48 bytes 48 bytes 48 bytes ATM header Cell payload SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 26

27 ATM for LANs H5 ATM links H4 Ethernet links H6 E1 E2 H3 Ethernet switch ATM-attached host H7 H1 E3 ATM switch H2 Attractive technology before switched LANs arrived High speed Switched technology can run long distances Difficult to make it behave like a LAN consider broadcasts needed to resolve addresses SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 27

28 LANE LAN Emulation Higher-layer protocols (IP, ARP,...) Signalling + LANE AAL5 Ethernet-like interface Higher-layer protocols (IP, ARP,...) Signalling + LANE AAL5 ATM ATM ATM PHY PHY PHY PHY Host Switch Host Provides a service interface that makes ATM circuit look Ethernet or token ring Main functions of LANE Resolve MAC addresses to ATM address Support broadcasts Problem how to emulation a shared medium using a virtual circuit network How is this different from switched LANs? SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 28

29 LANE LES BUS ATM network Point-to-point VC Point-to-multipoint VC H1 H2 Simple centralized protocol Clients register with a centralized server and use it for translation (using a well known address) A Broadcast server implements broadcasts To communicate address with a host with unknown Sent a packet to LES and first packet to BUS LES responds with ATM address, BUS broadcasts the packet VC can now be established, further packets through VC SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 29

30 Internetworking We know how to build directly connected networks Switches have the potential of scaling to large networks Bridges (link-level switching) provided valuable lessons Not scalable (consider spanning tree, or addressing scheme) Not heterogeneous (same MAC address family; compatible payloads, etc..) Scalability: must scale indefinitely Addressing and Routing Multicast and broadcast? Heterogeneity: Users on different networks must be able to talk Might need to cross several other networks on the way SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 30

31 Internet Protocol (IP) H1 H8 TCP R1 R2 R3 TCP IP IP IP IP IP ETH ETH FDDI FDDI PPP PPP ETH ETH Internet Protocol (IP; due to Karn and Cerf) Runs on all hosts (remember the hour-glass shape of the Internet Protocol Suite) Provides isolation from the networking technology Must provide one service model that is common to all possible underlying technologies what should we use? Connectionless best-effort delivery SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 31

32 IP Packet Format Version HLen TOS Length Ident Flags Offset TTL Protocol Checksum SourceAddr DestinationAddr Options (variable) Data Pad (variable) Version: 4 for IPv4 Hlen: number of 32-bit words in the header TOS: Type of Service; can be used for QoS Length: Number of bytes in the datagram Ident/Flags/Offset: used for fragmentation and reassembly TTL: Time to Live (maximum hop count allowed; why?) Protocol: key to identify higher level protocol (e.g., TCP) Checksum: applied to header (why not CRC?) Source and destination addresses (what addresses?) Options Payload Where does the link-layer header fit? SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 32

33 Fragmentation and Reassembly Why? H1 R1 R2 R3 H8 ETH IP (1400) FDDI IP (1400) PPP IP (512) PPP IP (512) PPP IP (376) ETH IP (512) ETH IP (512) ETH IP (376) Why is this needed? Each network has some Maximum Transmission Unit (e.g., ethernet 1500 bytes; PPP 532 bytes) Restrict IP payload to the smallest MTU; or Use fragmentation and reassembly if necessary Strategy: Fragment when necessary (MTU < Datagram) Fragments are self-contained IP packets Try to avoid fragmentation at the source Refragmentation is possible Delay reassembly until destination What if a fragment is lost? SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 33

34 Fragmentation and Reassembly How? Start of header (a) Ident = x 0 Offset = 0 Rest of header 1400 data bytes Start of header (b) Ident = x 1 Offset = 0 Rest of header 512 data bytes Ident = x Start of header 1 Rest of header 512 data bytes Offset = 512 Ident = x Start of header 0 Rest of header 376 data bytes Offset = 1024 Ident is the packet sequence number M-bit (flags field) is 1 in all but the last fragment How would refragmentation be implemented? path MTU discovery to minimize fragmentation SUNY-BINGHAMTON CS528 FALL 07 LEC. #9 34