File Servers

A file server is a combination of computer, internal hardware, and software that allows network users to share computer programs and data. A file server usually has a significantly faster processor, faster network interface card, more memory, and more data storage than most PCs. It may also have a tape back-up unit, modems, and several printers attached.

Server Hardware

Server software is that portion of the network operating system that "serves" a computer's resources to other network users. This software accepts incoming requests from network users and gives back files and other resources. In the days of MS-DOS, users achieved this connection by mapping a local logical drive, drive F: for instance, to the server's physical drive. That is, using a designated drive letter from A to Z, users access the server's physical drive as if it were one of their own local drives . Application programs do the same. The network operating system takes care of routing traffic across the network to the proper file server hard disk and back.

A file server must find data quickly and get to the requesting workstation with minimal delay. Factors affecting the server's performance include the CPU's speed, the network interface card's speed, the amount of RAM available, the type of disk and controller, the cable type and length, the network software's efficiency, the application type being run, and the number of users on the network.

Most high-performance file servers include a very fast, very large hard disk. Access time is the time required to get data off the disk, and it is broken into seek and rotation times. Average seek time, which accounts for the bulk of the time, is the time the disk head takes to move to the correct track on the platter. Average rotation time is the time taken for the platter to turn to the sector where the data is stored. Most of the time, the average access time is all you need to know when buying a server hard disk.

One overlooked aspect of the hard disk is its controller. Standard ST, Enhanced Small Device Interface (ESDI), Small Computer System Interface (SCSI), and Integrated Drive Electronics (IDE) are four types. ST drives are practically obsolete, ESDI drives are faster, but less common with each passing month, SCSI drives are easily expandable, and IDE drives are often the default on workstations and low-end server models. The tradeoff depends on your application and financial situation. Today, most network drives are SCSI.

Disk capacity is crucial. A rule of thumb is to figure out how much disk space you will need, and then double it. LANs always grow, so expandability is important. The file server should be able to accommodate more disk drives as needed. A file server (and a NOS) should permit you to add storage beyond any practical limit you can foresee. Disk arrays, which are composed of multiple drives and controllers, will provide fault tolerance but won't necessarily improve performance.

Memory Needs

A cache can make the hard disk appear to work faster than it actually does. Cache is space in high speed memory that is set aside to hold the last data read from the disk. The server software and the disk controller take more data off the disk than the user actually requested . This adjacent data, stored in the cache, is available when the next request is made, saving the file server from going to the disk to get it. It doesn't speed the work of the disk, but it eliminates some disk access, which moves data faster.

Caching is effective because a file server usually makes several accesses to disk when retrieving or writing data. Caching works because the next information a user requests is generally stored sectors adjacent to the data just requested. Caching also works because the disk controller usually can't get all the data with a single access to disk. So the disk controller reads the next couple of sectors after what the user requested, because chances are that the data the user wants next will be in those blocks.

Some server software caches only disk reads. This way, if the file server crashes, no data is lost because the data in the cache is identical to what is on the disk. Other file server software caches both read and write.

The amount of RAM necessary for disk caching varies from vendor to vendor. In general, a bigger cache is better, although a point of diminishing returns sets in. Moreover, the more data kept in a cache, the more it is vulnerable to server failure. A power loss will wipe out data in the cache, although data written to disk will not be lost. Most vendors take precautions for this, writing and verifying data at specified intervals, thereby protecting as much data as possible while still increasing performance.

Processor, Clock, And Ports

File servers should be more powerful than workstations for the shared services they provide. In some small networks with nondedicated servers, a last-generation processor will perform fine. However, once data sharing passes beyond a handful of users or light word processing, more powerful file servers are a necessity. Servers that provide only file and print services are likely to be more dependent on the amount of RAM, the disk subsystem performance, and the performance of the system bus than they are on the processor itself. Applications servers, on the other hand, can often absorb all the CPU processing power they can get.

Every server should have plenty of expansion slots as well as multiple serial and parallel ports. More than one or two locally connected printers can take up all available ports, since few PCs have more than one of each type of interface on the motherboard. Keep this in mind if you are using the file server as a router or switch. Two or three network interface cards, one or two drive controllers, and extra serial or parallel ports may occupy more ports than are available in the file server.

After performance, the main concerns when choosing a file server are reliability and compatibility. A PC file server needs to have a good power supply so it does not experience power drops or outages. It must work with standard software drivers such as network drivers, disk drivers, and video drivers, which means its BIOS must be compatible with those of other major vendors. Most cheap IBM clones do not work well as file servers.

This tutorial, number 8, written by Aaron Brenner, was originally published in the March 1989 issue of LAN Magazine/Network Magazine.