Chapter 5. Making Happy Users

It is said that "a day that is without troubles is not fulfilling. Rather, give me a day of troubles well handled so that I can be content with my achievements."

In the world of computer networks, problems are as varied as the people who create them or experience them. The design of the network implemented in Chapter 4, "The 500-User Office" may create problems for some network users. The following lists some of the problems that may occur:

Caution

A significant number of network administrators have responded to the guidance given here. It should be noted that there are sites that have a single PDC for many hundreds of concurrent network clients. Network bandwidth, network bandwidth utilization, and server load are among the factors that determine the maximum number of Windows clients that can be served by a single domain controller (PDC or BDC) on a network segment. It is possible to operate with only a single PDC over a routed network. What is possible is not necessarily best practice. When Windows client network logons begin to fail with the message that the domain controller cannot be found or that the user account cannot be found (when you know it exists), that may be an indication that the domain controller is overloaded or network bandwidth is overloaded. The guidance given for PDC/BDC ratio to Windows clients is conservative and if followed will minimize problems but it is not absolute.

Users experiencing difficulty logging onto the network When a Windows client logs onto the network, many data packets are exchanged between the client and the server that is providing the network logon services. Each request between the client and the server must complete within a specific time limit. This is one of the primary factors that govern the installation of multiple domain controllers (usually called secondary or backup controllers). As a rough rule, there should be one such backup controller for every 30 to 150 clients. The actual limits are determined by network operational characteristics.

If the domain controller provides only network logon services and all file and print activity is handled by domain member servers, one domain controller per 150 clients on a single network segment may suffice. In any case, it is highly recommended to have a minimum of one domain controller (PDC or BDC) per network segment. It is better to have at least one BDC on the network segment that has a PDC. If the domain controller is also used as a file and print server, the number of clients it can service reliably is reduced, and generally for low powered hardware should not exceed 30 machines (Windows workstations plus domain member servers) per domain controller. Many sites are able to operate with more clients per domain controller, the number of clients that can be supported is limited by the CPU speed, memory and the workload on the Samba server as well as network bandwidth utilization.

Slow logons and log-offs Slow logons and log-offs may be caused by many factors that include:

• Excessive delays in the resolution of a NetBIOS name to its IP address. This may be observed when an overloaded domain controller is also the WINS server. Another cause may be the failure to use a WINS server (this assumes that there is a single network segment).

• Network traffic collisions due to overloading of the network segment. One shortterm workaround to this may be to replace network HUBs with Ethernet switches.

• Defective networking hardware. Over the past few years, we have seen on the Samba mailing list a significant increase in the number of problems that were traced to a defective network interface controller, a defective HUB or Ethernet switch, or defective cabling. In most cases, it was the erratic nature of the problem that ultimately pointed to the cause of the problem.

• Excessively large roaming profiles. This type of problem is typically the result of poor user education as well as poor network management. It can be avoided by users not storing huge quantities of email in MS Outlook PST files as well as by not storing files on the desktop. These are old bad habits that require much discipline and vigilance on the part of network management.

• You should verify that the Windows XP WebClient service is not running. The use of the WebClient service has been implicated in many Windows networkingrelated problems.

Loss of access to network drives and printer resources Loss of access to network resources during client operation may be caused by a number of factors, including:

• Network overload (typically indicated by a high network collision rate)

• Server overload

• Timeout causing the client to close a connection that is in use but has been latent (no traffic) for some time (5 minutes or more)

• Defective networking hardware

No matter what the cause, a sudden loss of access to network resources can result in BSOD (blue screen of death) situations that necessitate rebooting of the client workstation. In the case of a mild problem, retrying to access the network drive of the printer may restore operations, but in any case this is a serious problem that may lead to the next problem, data corruption.

Potential data corruption Data corruption is one of the most serious problems. It leads to uncertainty, anger, and frustration, and generally precipitates immediate corrective demands. Management response to this type of problem may be rational, as well as highly irrational. There have been cases where management has fired network staff for permitting this situation to occur without immediate correction. There have been situations where perfectly functional hardware was thrown out and replaced, only to find the problem caused by a low-cost network hardware item. There have been cases where server operating systems were replaced, or where Samba was updated, only to later isolate the problem due to defective client software.

In this chapter, you can work through a number of measures that significantly arm you to anticipate and combat network performance issues. You can work through complex and thorny methods to improve the reliability of your network environment, but be warned that all such steps demand the price of complexity.