Global access to information is a requirement for corporations wishing to compete successfully in today's global economy. International corporate expansion through growth, mergers, partnerships, etc. has given rise to the disbursement of workgroups across multiple remote locations around the world. This has created an environment where file systems must span geographies. As a result, the definition of a network file server is no longer limited to a file server on the LAN. It has been expanded to include file servers over wide area networks (WANs) as well.

NFS supports global workgroups by giving users transparent, fast access to server file systems connected to client systems on LANs and WANs. The following capabilities enable efficient global file access:

• The automounter keeps files continuously accessible by automatically mounting file systems on servers located worldwide

• Client side caching gives clients fast access to file data transferred from the remote server and stored locally.

• To exchange information between the client and the server, NFS utilizes efficient transport protocols such as TCP [3]).

The following sections explain the NFS features that address file sharing over WANs.

Global, Transparent File Access

The automounter gives clients transparent access to server file systems, whether those servers are connected to clients over LANs and/or WANs. The automounter does not, nor does it need to, distinguish between different physical network technologies. It simply adds the server file system to the client file tree via the mount process, making the global network appear as if it is a single file system resource. The transparency offered by the automounter in combination with high performance file access gives global workgroups fast, easy access to file information.

Reducing Data Latency With Client Disk Caching

WANs are typically low bandwidth and prone to errors requiring data retransmission. This can result in significant data latency or in other words reduced throughput. NFS clients can offset data latency by utilizing local disk caching. By caching significant amounts of file data on the local disk, clients can drastically reduce the amount of time they would otherwise spend waiting for data to be transferred from the remote server. For more information on how local disk caching works in conjunction with NFS, refer to the section on "How Caching Increases NFS Performance" on page 11.

Efficient Data Exchange Using TCP

Because NFS is a layer 7 service, it relies upon lower layer transport protocols to transmit data over the network. NFS was originally designed to use UDP as the transport protocol for communication over both local and wide area networks. UDP was chosen because in the past, it provided better performance than TCP. But although UDP works very well for LANs, it has some limitations when used for communication over WANs. One drawback is that because it is an unreliable protocol, UDP is not designed to handle the special problems introduced by high latency, low bandwidth network connections.

Now that there are high performance TCP implementations available, NFS has been enhanced to utilize TCP. Unlike UDP, TCP is a reliable protocol that is designed to provide guaranteed data delivery. It independently performs tasks such as dynamic retransmission, packet sequencing, congestion control and error recovery, freeing NFS from having to cope with these problems. The result is a increase in NFS client performance.

Because there is more overhead required to set up an initial connection using TCP than there is using UDP, all NFS client traffic can be multiplexed over one TCP connection to the server. In other words, NFS clients use one connection to each server, regardless of the number of client mounts per server. This keeps overhead to a minimum, uses fewer resources (sockets, descriptors, etc.), and makes recovery from failure faster. It also allows servers to scale to support larger numbers of NFS clients.

How NFS Version 3 and TCP Work Together

NFS Version 3 works in conjunction with TCP to provide further performance gains. Unlike NFS Version 2, NFS Version 3 has no fixed limit on the amount of data that can be transferred between client and server in a single read or write request. The client and server are able to negotiate whatever transfer size they can both support. In general, the larger the transfer size, the more efficient TCP becomes because more data can be sent at one time.

Backward Compatibility with NFS UDP

Considering these benefits, it is easy to see why TCP is expected to become the preferred transport protocol for NFS, but it will take time for the installed base of UDP clients and servers to switch over. Therefore, in order to provide backward compatibility, NFS clients and servers can be implemented such that they can support both protocols. For example, when an NFS connection is in the process of being established on Solaris 2.5, the client and server will negotiate which protocol to use based on what they each support.Figure 2 shows how Solaris NFS clients and server can handle either protocol concurrently.

Figure 2 How Solaris NFS clients and servers support NFS TCP and NFS UDP concurrently.