• No se han encontrado resultados

A LAN is used to connect a number of computers and other devices in a small, local, area such as a room, a fl oor, or a building. A computer in a LAN could be a server, that is, it is permanently assigned a specifi c task (Web server, e-mail server, fi le server, or print server) or a client proper, which shares the fi les, printers, and servers with other client computers in the same LAN. Typically a hub or a switch is used to connect all the servers and clients with network cables. Each node in the LAN must have a Network Interface Card (NIC), which provides the physical layer and data link layer functions. Figure 3.1 shows a typical LAN and its primary components.

Each LAN also has a Network Operating System (NOS), which is distributed among the clients and servers. The NOS in a server handles all network functions, performs data link, network, and application layer functions, and acts as the applica- tion software by executing and responding to the requests sent to them by clients. MS Windows Server and Linux Server are two popular examples of a NOS for servers.

The NOS in a client computer provides data link and network layer functions and interacts with the application software and the computer’s own operating system. The client portion of the NOS software is typically included in the OS software such as Microsoft Windows, Vista, and Linux.

Ethernet, specifi ed in the IEEE 802.3 family of standards, is the most commonly used LAN technology today. There are two different implementation of Ethernet: traditional (or shared) and switched. The shared Ethernet architecture has a bus topol- ogy where all messages from any computer fl ow onto the central cable (bus) and any computer and device connected to this bus receives messages from all other comput- ers, whether the message is intended for it or not. When a frame is received by a computer, the fi rst task is to read the frame’s destination address to see if the message is meant for it or not. As discussed in later sections, Ethernet uses a technique, called Carrier Sense Multiple Access with Collision Detection (CSMA/CD), to regulate the sharing of the medium when multiple nodes try to transmit simultaneously. A hub, basically a dumb box containing the bus, is used to connect the computers on the shared Ethernet. Computers are linked into this central hub. If there are more devices to connect than the number of ports available on a hub, then another hub can be con- nected to the fi rst one in a cascade fashion to double-up capacity.

The switched Ethernet implementation uses a switch, instead of a hub. A switch is designed to support a small number of computers (16–24) in one LAN. Each com- puter is connected to the switch via a point-to-point circuit. A switch reads the desti- nation address of the incoming Layer 2 message (a.k.a., frame) and only sends it to the corresponding port (a hub broadcasts frames to all ports). To do this forwarding, the switch relies on a Forwarding Table, which has two columns: a column that contains the Layer 2 [also called Media Access Control (MAC)] addresses and another column containing the corresponding port numbers. The Forwarding Table is populated grad- ually. The switch starts with an empty Forwarding Table, working like a hub, and it gradually fi lls it by learning about the nodes that connect to it as follows: when the switch receives a frame from a port, it reads the source MAC address of the frame. If

Internet PC PC PC PC PC Server Printer Router/gateway Other LANs Router Hub/switch

the MAC address is not already in the table, the switch records it in a row that corre- sponds to the port number on the Forwarding Table. Then, the switch reads the desti- nation MAC address of this frame and checks whether the table already contains it. If the destination MAC address is in the table, the switch determines the corresponding port number from the table and forwards the frame to the computer that is connected to this port. If, however, the table does not include this MAC address, then the switch broadcasts this frame to all ports and waits for the destination computer to respond with an acknowledgment frame. Once the switch receives the response, it enters the source MAC address of this response frame in an appropriate row (i.e., the row that corresponds to the port where the frame came from). By continuing these steps, the switch fi lls all the entries in the table gradually in a relatively short time. The switched Ethernet is much faster and more reliable than the traditional Ethernet.

The cables connecting various components of a LAN are typically unshielded twisted pair wires categorized based on their quality. For example, Category 3 (Cat 3) and Cat 5 cables are suitable for low-speed Ethernet at 10–100 Mbps (Megabits per second) transmission rates. Faster Ethernets may require the use of fi ber optic cables carrying Gbps (Gigabits per second) data rates. By convention, 10Base-T for- mat refers to the Ethernet at 10 Mbps data rate with twisted pair cables. The word “Base” indicate that the transmission is a “base band” in a single channel covering the entire spectrum. A popular Ethernet technology currently commonly used is 100Base-T, which provides 100 Mbps data rate. 1000Base-T (or 1GBase-T) is also popular. More advanced Ethernet technologies typically use fi ber optic cables and are referred to as Gigabit Ethernets with GbE designation. See the last section of this chapter for more details on the GbE-related trends.

In putting together a LAN, perhaps the most important aspect to watch for is the effective data rate, which is the maximum speed in bits that the hardware layers (i.e., physical and data link layers) can provide. The effective data rate of a LAN has three major components: (i) the nominal data rate that depends on the physical layer prop- erties. For example, the nominal rate for a 100Base-T Ethernet is 10 Mbps; (ii) the error rate of the circuit, which is a critical factor in determining retransmissions, therefore reducing the transmission effi ciency; and (iii) the effi ciency of data link layer protocol. This is basically the percentage of transmission bandwidth that trans- ports user data, which depends on the number of overhead bits. (The error rate and the protocol effi ciency topics will be discussed later in the protocol section of the chapter.) The packet size, which plays an important role in these calculations, depends on the type of the traffi c on the LAN. A typical LAN traffi c includes a number of small Web application specifi c (i.e., HyperText Transfer Protocol, HTTP) or the e-mail application-specifi c (i.e., SMTP, or Simple Mail Transfer Protocol) messages (a.k.a., packets) followed by a large number of larger packets. This results in about 97% effi ciency for the Ethernet frames with 33 bytes overhead. To calculate the effective data rate for a whole LAN, we need to consider the effi ciency of the media access control. The Ethernet’s media access mechanism works well in low-traffi c LANs, resulting typically in a 50% capacity utilization. With these numbers, the total effec- tive data rate for a 100Base-T can be determined as follows: 50% capacity × 97% effi ciency × 100 Mbps rate = 485 Mbps (shared by all computers on the LAN). With 10 computers in the LAN, each computer’s effective data rate would be

485/10 = 48.5 Mbps. The effective rates for switched Ethernets are much higher due to the signifi cant improvements (up to 95%) in capacity utilization since there is no sharing in this architecture. Keep in mind that the switched Ethernet is not affected by the traffi c due to the availability of dedicated circuits for each computer.

Once the LAN is set up and operational, there needs to be continuous monitor- ing and, if necessary, to make changes to improve performance. In this regard, the throughput is an indicator of the performance. The name of the game in this per- formance improvement business is to identify and eliminate the bottlenecks, which are the components that can not handle all of the incoming traffi c. The identifi ca- tion of network bottlenecks is not that diffi cult. There are mainly two culprits: a server or a circuit. A key indicator to determine whether the server is a bottleneck is the utilization of the server. If a server is overutilized, say over 70%, then it is most likely that this server is the bottleneck. To eliminate this bottleneck situation, the performance of the server needs to be improved by upgrading the hardware (CPU, memory, storage, etc.), software (operating system, applications, etc.), or both. If the server utilization is low, then the bottleneck is the circuit. Improving the circuit capacity, say from 100Base-T to 1000Base-T may do the trick. Keep in mind that there may be other techniques to improve the performance. For example, dividing a LAN into two or more segments, reducing the network demand by scheduling certain applications during off hours, will defi nitely help to improve the performance of a LAN.

Documento similar