Ethernet Hub. Campus Network Design. Hubs. Sending and receiving Ethernet frames via a hub

Transcription

1 Campus Network Design Thana Hongsuwan Ethernet Hub 2003, Cisco Systems, Inc. All rights reserved , Cisco Systems, Inc. All rights reserved. BCMSN v Sending and receiving Ethernet frames via a hub Hubs? 5555 So, what does a hub do when it receives information? Remember, a hub is nothing more than a multi-port repeater. Nope 5555 Nope The hub will flood it out all ports except for the incoming port. Hub is a layer 1 device. A hub does NOT look at layer 2 addresses, so it is fast in transmitting data. Disadvantage with hubs: A hub or series of hubs is a single collision domain. A collision will occur if any two or more devices transmit at the same time within the collision domain. More on this later. 2003, Cisco Systems, Inc. All rights reserved. BCMSN v For me! Nope 2003, Cisco Systems, Inc. All rights reserved. BCMSN v

2 Hubs For me! Another disadvantage with hubs is that is take up unnecessary bandwidth on other links. Bridge 5555 Nope Wasted bandwidth Nope Nope 2003, Cisco Systems, Inc. All rights reserved. BCMSN v , Cisco Systems, Inc. All rights reserved. BCMSN v Traffic and LAN Size Bridged LAN Segments A: total traffic 6 Mbps B: total traffic 5 Mbps A sw/ B: traffic 7 Mbps Bridge B sw/ A: traffic 6 Mbps Joining LAN segments with a hub or repeater increases traffic. All machines share the same media (same collision domain). A+B: total traffic 11 Mbps repeater A bridge only forwards packets when necessary. learns network addresses of machines connected to each port doesn t forward traffic between machines on same port provides packet buffering and retiming, reducing collisions does forward all broadcast traffic may forward multicast traffic 2003, Cisco Systems, Inc. All rights reserved. BCMSN v , Cisco Systems, Inc. All rights reserved. BCMSN v

3 Transparent Bridging Types of Bridges Transparent bridges were first developed at Digital Equipment Corporation (Digital) in the early 1980s. Work into the IEEE standard. Transparent bridges are very popular in Ethernet/IEEE networks. When transparent bridges are powered on, they learn the workstation locations by analyzing the source address of incoming frames from all attached networks. If a bridge sees a frame arrive on port 1 from Host A, the bridge concludes that Host A can be reached through the segment connected to port 1. Through this process, transparent bridges build a table Transparent Bridge Remote bridge with identical data link protocol Can support different physical media Translating Bridge Connection with different data link protocol Frame conversion For example, Ethernet to Token ring or Token ring to Ethernet May require assembly and reassembly Transmission rate conversion 2003, Cisco Systems, Inc. All rights reserved. BCMSN v , Cisco Systems, Inc. All rights reserved. BCMSN v Bridging and Switching Why Bridging and Switching? Bridges forward traffic based on MAC level address A bridge may perform protocol conversion or speed matching between different LAN types Bridges provides buffering of packets A is a bridge with all ports use the same frame type; also called a LAN to distinguish from an ATM or telecommunications Decrease traffic on LAN segments Extend LAN without increasing congestion Bridge different network protocols Speed matching Security Reliability: fault isolation and bandwidth balancing 2003, Cisco Systems, Inc. All rights reserved. BCMSN v , Cisco Systems, Inc. All rights reserved. BCMSN v

4 LAN Extension Bridging Different Protocols Fast Ethernet 300 m 300 m Fast Ethernet Ethernet bridge Token Ring a can extend length limit of network, since it provides packet retiming and retransmission bridge: different media and protocol to extend length limit es are not subject to repeater count limit on ethernet a bridge can convert frame formats requires compatible network addresses, e.g. ethernet & token ring are OK, but not ethernet and ATM frame conversion may lose some information about the frame 2003, Cisco Systems, Inc. All rights reserved. BCMSN v , Cisco Systems, Inc. All rights reserved. BCMSN v Protocol Conversion Problems Speed Matching Ethernet Frame 1 byte Start delimiter 1 byte SOH Token Ring Frame 1 Access control Destination Source Frame length Data address address CRC 1 Frame control Destination Source Frame End Frame Data address address CRC delimiter status Ethernet does not have frame address seen or copied bits (set by receiver in Token Ring Frame Status byte) Ethernet does not have priorities or access control flags Token Ring frame may be too long for ethernet Maximum throughput of ethernet and token ring not the same: some frames may be dropped Token Ring doesn t have a length field: bridge must buffer and compute Fast Ethernet hub 100 Mbps ethernet Gigabit Ethernet server Fast Ethernet server 10 Mbps Ethernet hub 10 Mbps ethernet A can connect segments operating at different speeds How to handle overload of a slow or busy segment? back pressure (false collisions) drop frames Ethernet es can support 10, 100 Mbps, and gigabit Bridging ethernets is simple: packet formats are the same 2003, Cisco Systems, Inc. All rights reserved. BCMSN v , Cisco Systems, Inc. All rights reserved. BCMSN v

5 Security Ethernet Switches and Bridges Hub: shared media access Switch: selective access ooooh.. rats! A hub forwards all packets to all ports. Any host can listen to packets to/from another host, using programs like tcpdump, etherwatch, or snoop. A only forwards packets to port containing the destination host. Computers on other ports cannot eavesdrop. Address learning Forward/filter decision Loop avoidance 2003, Cisco Systems, Inc. All rights reserved. BCMSN v , Cisco Systems, Inc. All rights reserved. BCMSN v Address Learning Abbreviated MAC addresses Switch operate in promiscuous mode Switches are also known as learning bridges or learning es. A has a source address table in cache (RAM) where it stores source MAC addresses after it learns about them. A receives an Ethernet frame and searches the source address table for the Destination MAC address. If it finds a match, it filters the frame by only sending it out that port. If there is not a match if floods it out all ports. 2003, Cisco Systems, Inc. All rights reserved. BCMSN v No Destination Address in table, Flood 1 Abbreviated MAC addresses How does it learn source MAC addresses? First, the will see if the SA () is in it s table. If it is, it resets the timer (more in a moment). If it is NOT in the table it adds it, with the port number. Next, in our scenario, the will flood the frame out all other ports, because the DA is not in the source address table. 2003, Cisco Systems, Inc. All rights reserved. BCMSN v

6 Destination Address in table, Filter Destination Address in table, Filter Abbreviate d MAC addresses Most communications involve some sort of client-server relationship or exchange of information. Now sends data back to. The sees if it has the SA stored. It does NOT so it adds it. (This will help next time sends to.) Next, it checks the DA and in our case it can filter the frame, by sending it only out port 1. Abbreviated MAC addresses Now, because both MAC addresses are in the s table, any information exchanged between and can be sent (filtered) out the appropriate port. What happens when two devices send to same destination? What if this was a hub? Where is (are) the collision domain(s) in this example? 2003, Cisco Systems, Inc. All rights reserved. BCMSN v , Cisco Systems, Inc. All rights reserved. BCMSN v No Collisions in Switch, Buffering Collision Domains Collision Domains Abbreviated MAC addresses Unlike a hub, a collision does NOT occur, which would cause the two PCs to have to retransmit the frames. Instead the buffers the frames and sends them out port #6 one at a time. The sending PCs have no idea that there was another PC that wanted to send to the same destination. 2003, Cisco Systems, Inc. All rights reserved. BCMSN v Abbreviated MAC addresses When there is only one device on a port, the collision domain is only between the PC and the. (microsegment) With a full-duplex PC and port, there will be no collision, since the devices and the medium can send and receive at the same time. 2003, Cisco Systems, Inc. All rights reserved. BCMSN v

7 Other Information How long are addresses kept in the? 5 minutes is common on most vendor es. How do computers know the Destination MAC address? ARP Caches and ARP Requests Abbreviated MAC addresses How many addresses can be kept in the table? Depends on the size of the cache, but 1,024 addresses is common. What about Layer 2 broadcasts? Layer 2 broadcasts (DA = all 1 s) is flooded out all ports. 2003, Cisco Systems, Inc. All rights reserved. BCMSN v What happens here? Notice the Source Address Table has multiple entries for port # , Cisco Systems, Inc. All rights reserved. BCMSN v What happens here? What happens here? The filters the frame out port #1. But the hub is only a layer 1 device, so it floods it out all ports. Where is the collision domain? Collision Domain , Cisco Systems, Inc. All rights reserved. BCMSN v , Cisco Systems, Inc. All rights reserved. BCMSN v

8 Mac Address Table Table Aging Mechanism Much of the performance of a bridge is depend on the design of the address table search algorithm. How big should the address table be? If the table is full, what should I do with new potential entries? 2003, Cisco Systems, Inc. All rights reserved. BCMSN v , Cisco Systems, Inc. All rights reserved. BCMSN v What is CAM? Can this method of learning addresses be extended to a network of bridges? Content Addressable Memory is a special kind of memory! Read operation in traditional memory: Input is address location of the content that we are interested in it. Output is the content of that address. In CAM it is the reverse: Input is associated with something stored in the memory. Output is location where the associated content is stored. Q. How does B2 see the network? A port 1 port 2 B1 X B port 1 port 2 B2 Y C Z 2003, Cisco Systems, Inc. All rights reserved. BCMSN v , Cisco Systems, Inc. All rights reserved. BCMSN v

9 Can this method of learning addresses be extended to a network of bridges? Q. How does B2 see the network? A. B2 sees that A, X, B and Y are on port 1 (B1 is transparent!) Its forwarding table is A 1 B 1 C 2 X 1 Y 1 Z 2 port 1 port 2 B2 Ethernet latency Latency is the period of time from when the beginning of a frame enters to when the end of the frame exits the. Latency is directly related to the configured ing process and volume of traffic. A X B Y C Z 2003, Cisco Systems, Inc. All rights reserved. BCMSN v B2 Abbreviated MAC addresses 2003, Cisco Systems, Inc. All rights reserved. BCMSN v Memory buffering Memory Buffering Abbreviated MAC addresses An Ethernet may use a buffering technique to store and forward frames. Buffering may also be used when the destination port is busy. The area of memory where the stores the data is called the memory buffer. This memory buffer can use two methods for forwarding frame: port-based memory buffering shared memory buffering In port-based memory buffering frames are stored in queues that are linked to specific incoming ports. Shared memory buffering deposits all frames into a common memory buffer which all the ports on the share. 2003, Cisco Systems, Inc. All rights reserved. BCMSN v , Cisco Systems, Inc. All rights reserved. BCMSN v

10 Two ing methods Transmitting Frames Cut-Through Switch checks destination address and immediately begins forwarding frame. Store and Forward Complete frame is received and checked before forwarding. Store-and-forward The entire frame is received before any forwarding takes place. The destination and source addresses are read and filters are applied before the frame is forwarded. CRC Check done Cut-through The frame is forwarded through the before the entire frame is received. This mode decreases the latency of the transmission, but also reduces error detection. Depends on the model of the. Fragment-Free Switch checks the first 64 bytes, then immediately begins forwarding frame. 2003, Cisco Systems, Inc. All rights reserved. BCMSN v , Cisco Systems, Inc. All rights reserved. BCMSN v Cut-through Cut-through : Fragment Free Cut-through Fast-forward Offers the lowest level of latency. Fast-forward ing immediately forwards a packet after reading the destination address. There may be times when packets are relayed with errors. Although this occurs infrequently and the destination network adapter will discard the faulty packet upon receipt. Cut-through Fragment-free Fragment-free ing filters out collision fragments before forwarding begins. In a properly functioning network, collision fragments must be smaller than 64 bytes. Anything greater than 64 bytes is a valid packet and is usually received without error. Fragment-free ing waits until the packet is determined not to be a collision fragment before forwarding. 2003, Cisco Systems, Inc. All rights reserved. BCMSN v , Cisco Systems, Inc. All rights reserved. BCMSN v

11 Bridge performance Ethernet Hubs vs. Ethernet Switches A bridge performs two performance-intensive operations: It implements the decision-making algorithms table lookup, filter/forward decision, and table maintenance. The worst case situation : every port bombarded with the maximum number of frame. This generally occurs with minimum length frames. (64 bytes) How maximum rate of forwarding for 10, 100,1000 Mb/s Bridge that can performs, call Wire Speed It move data among its ports. (data flow) From data flow perspective, the worst case is to have the longest possible frames, Bridge with 24 ports of 100 Mb/s = 24 x 100 Mb/s = 2.4 Gb/s Bridge that can performs, call Non Blocking An Ethernet is a packet for Ethernet frames Buffering of frames prevents collisions. Each port is isolated and builds its own collision domain An Ethernet Hub does not perform buffering: Collisions occur if two frames arrive at the same time. Hub Switch Input Buffers HighSpeed Backplane Output Buffers 2003, Cisco Systems, Inc. All rights reserved. BCMSN v , Cisco Systems, Inc. All rights reserved. BCMSN v Ethernet Hubs vs. Ethernet Switches Full- and Half-Duplex Full- and Half-Duplex Half-duplex: send or receive, one at a time. Full-duplex: send and receive simultaneously. 10 Mbps Ethernet supports half-duplex; fullduplex is not consistently implemented. 100 Mbps supports half- and full-duplex. A repeater can only operate in half-duplex mode. 2003, Cisco Systems, Inc. All rights reserved. BCMSN v , Cisco Systems, Inc. All rights reserved. BCMSN v

12 Ethernet Duplex Mismatch Half-Duplex Design One of the most common causes of performance issues on 10/100Mb Ethernet links is when one port on the link is operating at half-duplex while the other port is operating at full-duplex. If one end of a connection (device or Ethernet ) is set for auto-negotiation, and fails to see auto-negotiation at the other end, the former sets itself to the default, half-duplex. Auto-negotiation can sometimes fail, even when both sides are set to auto (although this isn t as prevalent as in the past). Ethernet Controller Collision Detection Ethernet NIC Tx Loopback Rx Transmit Receive Tx Loopback Rx Collision Detection Ethernet NIC Ethernet Controller Ethernet physical connector provides several circuits Most important are TX (transmit), RX (receive), and CD (collision detection) 2003, Cisco Systems, Inc. All rights reserved. BCMSN v , Cisco Systems, Inc. All rights reserved. BCMSN v Full-Duplex Ethernet Design Auto-Negotiation Eases Configuration Full Duplex Ethernet Controller Collision Detection Tx Loopback Rx TX RX Loopback Full Duplex Ethernet Controller Transmit circuit connects directly to receive circuit No collisions Significant performance improvement Eliminates contention on Ethernet point-to-point links Uses a single port for each full-duplex connection Tx Rx Collision Detection Starting from Fast Ethernet For UTP cables Auto negotiation about duplex mode (half or full), maximum speed (10 or 100 Mbps), and flow control support. Avoid configuration headache. The same RJ45 jack can support 10 or 100Mbps and half or full duplex mode, how do you know which setting you should use when you plug your RJ45 into a hub or a s port? So, the best setting that you should choose is autonegotiation. 2003, Cisco Systems, Inc. All rights reserved. BCMSN v , Cisco Systems, Inc. All rights reserved. BCMSN v

13 Auto-Negotiation Message Format Auto-Negotiation Message Transfer Because link speed is one of the parameter that needs to negotiate, the auto-negotiation message cannot be transferred at a chosen speed (e.g., 10 or 100 Mbps). Instead, it is transmitted at a predetermined clock rate. 2003, Cisco Systems, Inc. All rights reserved. BCMSN v , Cisco Systems, Inc. All rights reserved. BCMSN v Switch Congestion Flow Controls Needed When Ethernets Switches Are Used When a hub is used to connect all sending hosts (halfduplex mode), Ethernet s MAC algorithm is an effective flow/congestion control (backoff) and retransmission (up to 16 times) method. However, when a is used and full-duplex mode is used, each host now has its own collision domain and no longer works. Frames now may be lost due to buffer overflow inside the. Hosts no longer know this type of collision. Congestion control cannot automatically be performed. Also, retransmission cannot automatically be performed. 2003, Cisco Systems, Inc. All rights reserved. BCMSN v , Cisco Systems, Inc. All rights reserved. BCMSN v

14 Backpressure Can Be Used In Half- Duplex Mode IEEE 802.3x Flow Control If a uses half-duplex mode to connect to hosts, although each host has its own collision domain, we can use some methods to ask sending hosts to slow down their sending rates: Force collisions with incoming frames Make it appear as if the channel is busy If a uses full-duplex mode, the above two methods won t work. We need an explicit flow control for ed full-duplex Ethernet es. Introduced for Fast and Gigabit Ethernets. Introduce special MAC-layer control packets (PAUSE packets) to enable or disable frame transmission. Thus a more generic MAC framework is defined and an entity called (MAC Control) is introduced. MAC Control is responsible for generating, sending, receiving, and performing the PAUSE operation. 2003, Cisco Systems, Inc. All rights reserved. BCMSN v , Cisco Systems, Inc. All rights reserved. BCMSN v IEEE 802.3x Architecture MAC Control Frame Format 2003, Cisco Systems, Inc. All rights reserved. BCMSN v , Cisco Systems, Inc. All rights reserved. BCMSN v

15 PAUSE Frame Pause Time Pause_time field: number of pause-quanta (from 0 to 65535) which indicate the pause time pause-quanta = 512 bit time speed equal or less than 100 Mb/s T-Pause in bit time = pause-quanta * 512 Speed greather than 100 Mb/s T-Pause in bit time = pause-quanta * 512 * , Cisco Systems, Inc. All rights reserved. BCMSN v , Cisco Systems, Inc. All rights reserved. BCMSN v PAUSE Function PAUSE Frame Implementation Implement a simple stop-start flow control scheme. If a device wants to temporarily inhibit incoming frames, it sends a PAUSE frame to the full-duplex partner. This PAUSE frame contains a parameter indicating the length of time the partner should wait before sending more frames. If the device wants to cancel the timer at its partner which is set up by its previous PAUSE frame, the device can send another PAUSE frame that contains a parameter of zero time. PAUSE frames have higher priority over normal data frames. 2003, Cisco Systems, Inc. All rights reserved. BCMSN v , Cisco Systems, Inc. All rights reserved. BCMSN v

16 Must Stop Transmission in 512 Bit Time Buffer Thresholds to Control When to Issue PAUSE Frames After receiving a PAUSE frame, the receiver has 512 bit time to decode it. No more new transmission is allowed after this 512 bit time period. The buffer space (above the high watermark and below the low watermark) should be larger than link RTT * link BW. Otherwise, frames may be dropped or link utilization may not be 100%. 2003, Cisco Systems, Inc. All rights reserved. BCMSN v , Cisco Systems, Inc. All rights reserved. BCMSN v Input-Buffered Switches Assumed for IEEE 802.3x IEEE 802.3x is a linklevel flow control scheme. It uses the buffer occupancy level of an input port to determine whether to inhibit frames from incoming. Unfortunately, this scheme does not work for output-buffered. (Why? because there will be no queue in input ports.) IEEE 802.3x May Cause Low Link Utilization IEEE 802.3x is a link-level flow control applied to input-buffered es. If the input-buffered has the HOL blocking problem, then using IEEE 802.3x will not further lower the link utilization. (It is already low.) However, if the input port uses virtual output queuing (one queue for each output port) to avoid the HOL blocking problem, using IEEE 802.3x may further lower the link utilization. 2003, Cisco Systems, Inc. All rights reserved. BCMSN v , Cisco Systems, Inc. All rights reserved. BCMSN v

17 Low Link Utilization Example Basic Switch Specification (TL-SG1016D) Output port 1 is busy or blocked all the time. Switch B Switch A P1, P1, P1, P1 Output port 2 is idle. PAUSE P1 P2 P2 P2 After receiving the PAUSE frame, A cannot send any more frame to B. However, sending P2 should be allowed because P2 can be immediately forwarded when it arrives at B. 2003, Cisco Systems, Inc. All rights reserved. BCMSN v , Cisco Systems, Inc. All rights reserved. BCMSN v Switch Specification (TL-SG1016D) 2003, Cisco Systems, Inc. All rights reserved. BCMSN v , Cisco Systems, Inc. All rights reserved. BCMSN v

18 Cisco 100 Series 2003, Cisco Systems, Inc. All rights reserved. BCMSN v , Cisco Systems, Inc. All rights reserved. BCMSN v Cisco 100 Series Cisco 100 Series 2003, Cisco Systems, Inc. All rights reserved. BCMSN v , Cisco Systems, Inc. All rights reserved. BCMSN v

19 End. 2003, Cisco Systems, Inc. All rights reserved. 1-73