Network performance tuning

Table 1 contains network performance tuning information.

Table 1. Network performance tuning symptom and action information
Symptom	Action
Data transfer rates significantly less than network capacity	ACTION 1: Know your network topology - What is the NFS environment? briefly introduces the role of the network in the NFS environment. Clearly, transferring data over congested networks results in poor performance. Therefore, when a performance problem is encountered in the NFS environment, it is useful to determine whether or not the network is the source of the problem prior to investigating other alternatives. While there are products specifically designed to monitor and report on network activity, knowing your particular network topology may be sufficient for initial tuning. Figure 1 shows a simple network topology with two 100 Mbps Fast Ethernet Rings connected to a Gbps Fast Ethernet backbone network. In this example, access from NFS client A, on a 100 Mbps Fast Ethernet Ring, to NFS server B, directly connected to the Gbps Fast Ethernet backbone, would be as limited by network bandwidth as access to NFS server A. On the other hand, access from NFS client B, directly connected to the Gbps Fast Ethernet backbone, would be limited by the load on the Gbps Fast Ethernet network and the capacity of NFS server B.

Figure 1. Sample network topology

Factors to consider in addition to the theoretical capacity of a network configuration

There are other factors to consider in addition to the theoretical capacity of a network configuration. For instance, there are differences in the capability of network adaptors and network controllers. Overall network performance can also be influenced by the devices connecting network segments or subnets, particularly in terms of packet size limits and UDP checksum processing. While it may initially appear that there is no network constraint, there may be a device, adaptor, or network segment that is not performing at expected levels.

With this type of investigation, you may already have discovered that your primary constraint is a congested network. If so, you may elect to pursue such alternatives as:

Increasing network bandwidth
Changing network topology
Scheduling applications during low system usage
Modifying applications to reduce data transfer

ACTION 2: Monitor network activity - It may be necessary to monitor network activity over a period of time to determine what is causing a performance problem. In addition to whatever monitoring your network administrator may be able to provide, the netstat -s command, available on many NFS client platforms, may provide some insight as well. This command reports such statistics as bad checksums, dropped fragments, non-forwardable packets, various time-outs, and socket buffer overflows. In some cases, it may be necessary to use a network analyzer to determine exactly what is being sent over a network segment. While it can be difficult to analyze congested networks, network analyzers occasionally capture evidence that may indicate a problem with the NFS client or server system.