CCNP Enterprise Certification Study Guide: Implementing and Operating Cisco Enterprise Network Core Technologies. Ben Piper
Читать онлайн книгу.
the behavior of a shared medium. Time for a quick history lesson.
A Brief History of Ethernet
The original Ethernet standards from the 1970s were designed for nodes all connected to a shared electrical bus that often took the form of a thick yellow cable (you may have heard the term Thicknet). Whenever one node would transmit a signal, all other nodes connected to the cable would receive it. Communication was half-duplex, and all nodes were in the same collision domain. As a way of detecting errors introduced by collisions, the original Ethernet II (DIX) specification got a frame check sequence (FCS, sometimes called a cyclic redundancy check, or CRC) to detect errors. Even today as back then, nodes discard frames that fail the FCS check.
The multi-access nature of Ethernet made it necessary to assign each node's network interface a unique, 48-bit Media Access Control (MAC) address. The sending node would construct a frame that included the destination node's MAC address and the data to send. All nodes would receive the frame, but only the destination node would process it.
Now let's fast-forward to today. We still use MAC addresses, even though the original rationale for using them is long gone. To maintain backward compatibility over the years, we never got rid of them. Figure 1.3 shows the original DIX frame format that we still use today. We're still using a technology designed specifically for devices that were all sharing a thick yellow cable. Today, however, instead of nodes sharing this thick yellow cable, they're connected to a switch.
Figure 1.3 Layer 2 frame and layer 1 packet, structurally identical to the revised (1997) IEEE 802.3 format that we use today
Switches replace the shared cable of the early Ethernet with multiple cables, breaking the inherent broadcast nature of the thick yellow cable. Switches thus have to perform some interesting hackery to maintain backward compatibility with the early Ethernet standards. When a switch receives an Ethernet frame, by default it forwards that frame to all other devices connected to the switch—a process called flooding or broadcasting. This creates the illusion that all nodes are connected to the same thick yellow cable. (In networking parlance, they're all in the same broadcast domain or segment.) This illusion is called transparent bridging (aka switching) because to the nodes, the switch is invisible. Incidentally, you'll recognize this simulated yellow cable by its common name: a local area network (LAN).
The MAC Address Table
Although switches eliminate collision domains by offering full-duplex communication, they still waste bandwidth by flooding traffic to nodes that don't need it. To mitigate flooding, switches implement a form of routing. When a switch receives a frame on an interface, it records the ingress interface and source MAC address in its MAC address table. Subsequently, when a switch receives a frame destined for that same MAC address, it queries the MAC address table, which returns the interface number. The switch then forwards the frame only out of that interface, rather than flooding it.
The MAC address table is stored in a type of memory called content-addressable memory (CAM). CAM is often used as a synonym for the MAC address table. The CAM takes a MAC address and VLAN as input and returns an interface name and number as the output. CAM provides faster read times than RAM.
SW3#show mac address-table dynamic Mac Address Table ------------------------------------------- Vlan Mac Address Type Ports ---- ----------- -------- ----- 1 0c3c.8a00.5e02 DYNAMIC Gi0/2 1 0c3c.8ad7.9101 DYNAMIC Gi0/2 1 0c3c.8afd.c101 DYNAMIC Gi0/1 1 0c3c.8afd.c102 DYNAMIC Gi0/2 10 0c3c.8ad7.800a DYNAMIC Gi0/0 20 0c3c.8ad7.8014 DYNAMIC Gi0/0 Total Mac Addresses for this criterion: 6
On the other hand, if a switch receives a frame for a MAC address that doesn't have a mapping in the MAC address table—called an unknown unicast—it reverts to its default behavior and floods the frame out of all other interfaces.
Unknown unicasts are more common than you might think. Entries in the MAC address table don't last forever. By default, a MAC address entry is deleted or ages out 300 seconds (5 minutes) after the switch last sees the traffic from the MAC address. Note that aging time is not based on when the entry was created.
SW3#show mac address-table aging-time vlan 1 Global Aging Time: 300 Vlan Aging Time ---- ---------- 1 300
You can adjust the global aging time to between 10 and 1,000,000 seconds or disable aging by setting the aging time to 0.
SW3(config)#mac address-table aging-time ? <0-0> Enter 0 to disable aging <10-1000000> Aging time in seconds
You can also adjust the aging time on a per-VLAN basis.
SW3(config)#mac address-table aging-time 300 vlan ? <1-4094> VLAN id
Disabling aging might sound like a good idea, as it would prevent flooding, right? Not necessarily. The CAM has a finite amount of space, and once the MAC address table is full, the switch will flood traffic to every destination MAC not in the table.
The MAC address table mitigates flooding but doesn't eliminate it. The fundamental flooding behavior of Ethernet remains. To make matters worse, Ethernet implements a special MAC address called a broadcast address (FFFF.FFFF.FFFF). Frames sent to this address are flooded out of all ports. You can imagine the number of major outages that arose from this unwise decision!
The end result is that any node in a broadcast domain can send a frame to another node and the destination node will receive it. We may have added more cables and more devices, but the fundamental behavior of Ethernet hasn't changed in 50 years, as shown in Figure 1.4. When it comes to networking, history has a way of repeating itself.
Figure 1.4 Early Ethernet over a shared medium compared to Ethernet using a switch
Maximum Transmission Unit
Another interesting side effect we inherited from the use of the legendary thick yellow cable is that Ethernet had to impose a limit on the maximum frame size to keep a single node from hogging the medium with colossal frames. The Ethernet maximum transmission unit (MTU) defines the maximum size of the Data field in bytes. DIX and IEEE 802.3 support a maximum MTU of 1,500 bytes. Higher-layer protocols trying to send packets larger than the MTU must break apart their packets into fragments that will fit into the frame's Data field. To avoid fragmentation, some interfaces support jumbo frames with an interface MTU of 9,000 bytes to 9,216 bytes.
Subnet Limits
When