An electronic filesystem employing the following technique:

Two block devices are available to filesystem, one on relatively cheap storage media with slow access times or throughput (such as magnetic disc, or "hard-drive"), one on more expensive media with faster seek times or throughput, such as raid flash storage.

The amount of "slow" media vs. "fast" media could be adjusted to suit the application. A starting ratio might be 60% of space on "slow" media, and 40% on "fast" media.

A global threshold exists in the filesystem for the number of blocks that a file may consume on the "fast" media before the file starts to utilize "slow" media.

The filesystem can be thought of as a combination of the two areas on the separate media; similar to Linux's LVM. Addressing is such that inodes can reside on either media, and a single block address specifies an exact block on either the "fast" or "slow" media. For example, if the "fast" media had only 10 blocks, and "slow" only 10 blocks, the complete filesystem would consist of 20 blocks, numbers 0-9 on the "fast" media, and numbers 10-19 on the "fast" media.

As files are created on the filesystem, they are always created on the "fast" media. As data is written, it is first written entirely to the "fast" media. As data is written to the media, the filesystem tracks how many blocks exist in the file. Once the number of blocks passes the global threshold, the filesystem switches to using the "slow" media for subsequent blocks. The remainder of the file will then be written to the "slow" media. Random-access makes this a bit more complicated, but basically, the file's low-numbered blocks (numbering blocks in the file, not the filesystem) can all be kept on the "fast" media.

If the amount of unused "fast" media becomes low, the global threshold can be adjusted down, to use more "slow" media. Conversely, if the amount of unused "slow" media becomes low, the global threshold can be adjusted up.

This system has the advantage of very fast access times for small files, many of which are accessed during system boot.

Data in larger files, if accessed sequentially such as multimedia files, will not suffer as much seek delay if the media is not heavily fragmented. Additionally, the filesystem can quickly get started with a sequential read of a large file, since the beginning resides on the "fast" media, and if prefetch is enabled, the head of the "slow" media can move to the first point it needs to read, while the computer application is already processing the beginning of the file.

This technique could be extended by hard-disk vendors by supplying both "fast" flash media, and "slow" magnetic-disk media in a single package, and passing information to the host specifying which blocks are on which media. This could allow computer operating systems to automatically utilize the technique, with little or no user intervention.

Licensed freely for anyone to use for any purpose other than terrorism.
Published to attempt to prevent others patenting.