Defragmenting helps disk performance

Fragmented files are the dark side of random access storage.

Files are fragmented every time you update or delete files or download something from the Web. That's because a file is rarely stored contiguously; pieces of it are randomly scattered - or fragmented - throughout the disk. Although this means that the disk's capacity will be used efficiently, it also slows I/O operations because the disk's read/write head has to pick up file fragments from many different locations to reassemble the file. Enough of this I/O latency can bring a desktop or server to its knees.

Many operating systems have defragmenters, ranging from rudimentary ones bundled with desktop operating systems to more comprehensive server-oriented products. Fundamentally, the defragmenters all work the same way: They temporarily move a file's pieces to blank areas of the disk and gradually reconstruct the pieces into a whole contiguous entity. Performance improves because the read/write head has to look in just one place to read the entire file.

Variety of files

A defragmenter has to deal with four types of files: an index file, paging files, directories and data files.

The index file - called the master file table in a Windows NT file system volume and the file allocation table in Windows 95 and DOS - is the most important of the files. It contains all the information an operating system needs about every file on the disk, including a file's name, size, location, time and date stamps, and permissions. The index file is the first place that a request for a file goes, so making the index file contiguous means that the files being requested will be found and accessed more quickly.

Paging files are spaces on the disk where data from the main memory is swapped in and out. In Windows 95 and 98 the spaces grow or contract as needed, but the files end up fragmented. This increases the time it takes to open and close files and applications. Defragmenting the paging file and putting it in one location also speeds performance.

Directories are catalogs of file names and other directories. They contain pointers to a file's location and tell the index file where to find a particular file. Like paging files, directories can become fragmented, which increases the time it takes to locate files.

The defragmenter finds the pieces of index files, paging files and directories, and reunites them to form contiguous entities.

The way defragmenters classify files and locate them on a disk varies by operating system and product. In general, files can be arranged as desired, such as alphabetically by file name or by their likelihood to be used. Statistically, files that have been used recently are the most likely to be used again, so the latter approach uses a file's time and date stamp to determine where the file will be placed on the disk. Recently used active files are placed on the middle part of the disk, between the spindle and the outer edge. Files located there are accessed faster than files in other areas because the read/write head has to travel a shorter distance to find and read them.

Two methods

It can take a long time to defragment a large and cluttered disk. Defragmenters use two methods to minimize their effect on operations. The first approach is to move only the files whose usage category has changed; only a portion of a disk's files has to be moved, so the defragmentation run will take less time. The second technique is to defragment open files so a disk can be defragmented while it is being used, instead of taking the system offline for defragmenting.

Defragmenting may be a complex operation internally, but user-friendly interfaces bring the process down to earth. Remote scheduling and control capabilities enable network administrators to defragment servers and desktop machines at any time from anywhere in the network. That ability gives net administrators a way to ensure that all their systems, wherever they are located, have orderly, structured disk files.

Williams is product manager at Raxco Software, a provider of disk utility software in Gaithersburg, Md. He can be reached at jwilliams@ raxco.com.