5 Network Hardware and Software

		layer usually provides error checking, TCP provides its own final data integrity checking. Most multiple-hop physical transports provide only a best-effort delivery promise. TCP incorporates a positive-acknowledgement sliding-window retransmission mechanism that recovers from packet loss. It also tolerates latency while maintaining high performance in the normal case of no packet loss. TCP provides its own data stream packetization, avoiding fragmentation in the IP layer.

		A second common service is UDP. UDP provides unsequenced, unreliable datagram transport. The advantages of UDP are that it has a relatively low latency because it incurs no start-up delay. Its primary disadvantage is that you typically have to provide retransmission services similar to those of TCP when you use UDP.

		The drawbacks of TCP come from its ability to handle wide-area networks. New TCP streams utilize "slow start" to gradually detect the bandwidth limit of the network. Congestion is detected by recording dropped packets. Any corrupted or dropped packet immediately drops the offered load. The Nagel algorithm, used by TCP can cause problems for message passing libraries. In order to minimize "tinygrams" (short packets), Nagel's algorithm delays the sending of small messages until the acknowledgment for an initial small message is returned. You can avoid this behavior by either compiling the Linux kernel with an option not to use Nagel's algorithm, or by constructing your programs to use large messages.

		5.3.1 IP Addresses

		The destination of an Internet Protocol packet is specified by a 32bit long IP address(or 128 bits for IPv6) that uniquely identifies the destination host. IP addresses are usually written in "dotted decimal notation," with the bytes of the address written as a decimal numbers separated by decimal points. The IP address range is divided into networks along an address bit boundary. The portion of the address that remains fixed within a network is called the network address and the remainder is the host address. The division between these two parts is specified by the netmask. A typical netmask is 255.255.255.0, which specifies a 24 bits of network address and eight bits of host addresses.

		Three IP address ranges have been reserved for private networks:

		10.0.0.0- 10.255.255.255

		172.16.0.0- 172.31.255.255

		192.168.0.0- 192.168.255.255

		These address ranges are permanently unassigned and will not be forwarded by Internet backbone routers or conflict with publicly accessible IP addresses. We