SC101/SC101T Defragmenting and Disk Maintenance

The SC101/SC101T is practically maintenance-free. It uses a SAN File System (SFS) that does not need you to defragment it, or to use other Windows utilities such as chkdsk. For the SC101/SC101T's own defragmenter to work properly — and as with any hard disk, to avoid poor performance — leave 10% free space on each disk or drive. (Remember, the SC101/SC101T uses the word drive to mean what is usually called a disk partition.)

Technical Explaination

With conventional disk management such the NTFS used by Windows, disks are divided into small, equally-size divisions called clusters; disk access is made for clusters, and a file may use many clusters. This old practice is suited for programs with limited needs for disk performance, or for low-level programs that handled their own disk access. Unfortunately, neither of these is true of many modern software applications.

To reduce request fragmentation for large requests, SFS bypasses the Windows buffer cache and tries to make one request for the entire amount of data.

The SFS accesses hard disks by addressing entire extents, which are flexibly-sized groups of blocks. With a good disk allocator such as ours, extent-based file systems minimize fragmentation. For example, say Windows makes 64 KB requests for buffered I/O, which results in eight I/O requests for 8 KB blocks. The SC101/SC101T SFS allocator, in comparison, searches chunks of thirty-two 8KB blocks (before moving to sixteen 8KB blocks, before finally moving to eight 8KB blocks). Often files are larger than the maximum Windows request size, so Windows’ files will fragment. SFS avoids the fragmentation of Windows. The SFS allocator can use both first-fit and best-fit algorithms, and typically outperforms Windows in terms of maximum blocks that can be transferred at a time. All this makes it easy for the SC101/SC101T to stay “defragmented”, using its own software.

Another non-Windows feature that reduces SC101/SC101T disk accesses is stuffed dinodes. In the Windows NTFS disk management, the addresses for blocks are stored in a different place than the blocks, themselves. This means that any request for NTFS disk data, no matter how small, causes two I/O operations. Since hard disk I/O is far slower than your CPU or RAM, doubling the amount of I/O can have a noticable impact on performance. Since SFS can store both the address and the data in the same extent, the amount of I/O for very small files is cut in half.

N101571.asp Oct. 24, 2005