UNIX System Administration Handbook - Evi Nemeth [163]
Maximum transfer units
The size of packets on a network may be limited both by hardware specifications and by protocol conventions. For example, the payload of an Ethernet frame can be no longer than 1,500 bytes. The size limit is associated with the link-layer protocol and is called the maximum transfer unit or MTU. Table 13.2 shows typical values for the MTU.
Table 13.2 MTUs for various types of network link layer
For ATM, the term MTU doesn’t quite apply; ATM lies somewhere in between the physical layer and the link layer. An ATM cell is normally 53 bytes with a 48-byte payload, but with AAL/5 framing, packets can be as large as 216. The ATM community typically chooses an MTU of 9,180 for normal mode and 1,500 for LANE (Local Area Network Emulation) mode.
In the TCP/IP suite, the IP layer splits packets to conform to the MTU of a particular network link. If a packet is routed through several networks, one of the intermediate networks may have a smaller MTU than the network of origin. In this case, the router that forwards the packet onto the small-MTU network will further subdivide the packet in a process called fragmentation. Fragmentation is an unwelcome chore for a busy router. The TCP protocol can determine the smallest MTU along the path to the destination and use that size from the outset. UDP is not so nice and is happy to shunt extra work to the IP layer. In the IPv6 protocol, intermediate routers can no longer perform fragmentation—MTU discovery is required.
Fragmentation problems can be insidious. If you are using a tunneled architecture for a virtual private network, for example, you should look at the size of the packets that are traversing the tunnel. They are often 1,500 bytes to start with, but once the tunneling header is added, they become 1,540 bytes or so and must be fragmented. Setting the MTU of the link to a smaller value will avert fragmentation and increase the performance of the network. Consult the ifconfig man page to see how to set an interface’s MTU.
Packet addressing
Like letters or email messages, network packets must be properly addressed in order to reach their destinations. Several addressing schemes are used in combination:
• MAC (media access control) addresses for hardware
• IP addresses for software
• Hostnames for people
A host’s network interface may have a link-layer MAC address that distinguishes it from other machines on the physical network, an IP address that identifies it on the global Internet, and a hostname that’s used by humans.
The lowest level of addressing is dictated by network hardware. For example, Ethernet devices are assigned a unique 6-byte hardware address at the time of manufacture. Token ring interfaces have a similar address that is also six bytes long. Some point-to-point networks (such as PPP, described on page 291) need no hardware addresses at all; the identity of the destination is specified as the link is established.
A six-byte Ethernet address is divided into two parts: the first three bytes identify the manufacturer of the hardware, and the last three bytes are a unique serial number that the manufacturer assigns. Sysadmins can often identify at least the brand of machine that is trashing the network by looking up the 3-byte identifier in a table of vendor IDs. A current vendor table is available from
http://www.iana.org/assignments/ethernet-numbers
This information used to be published regularly in the RFC series, but it is no longer distributed that way. RFC1700 (1994) was the last