Random versus Sequential I/O and How It Affects Your Data

Computer Data Network

Storage media have continuously evolved since Konrad Zuse used old movie reels for his automatic calculators in the 1930s. For the past several decades, hard disk (HDD) storage has reigned supreme. Companies have been able to cram more data onto smaller platters, ramping up storage density and, after a few years of high prices, dropping the cost per gigabyte to mere pennies. HDDs make our modern desktop computing world possible. If you want more storage for your music, photos and movies, you can just buy a larger hard drive and pop it into your computer.

That’s not quite the case for servers, though. Sure, you can install a new network-attached storage or storage area network and increase your storage pool, but you still have to pay a significant cost, although not in terms of money. In relying on hard disk storage, many companies trade off better performance and have serious issues with access times.

Sequential Access

Sequential input-output (I/O) is a form of I/O in which data is read from a storage medium in the same order it was stored. Hard disks are great for tasks that use storage sequentially, such as a database saving new entries or transferring large files from one drive to another, because they rely on positioning read-write heads over spinning platters.

Storing data in sequence is the most efficient use of a hard drive, because the platters rotate and the read-write heads float over them like the needle on a record player, though they never actually touch. If the next file begins immediately after the previous one ends, that head doesn’t have to move far.

Random Access

Computer Data NetworkWhere hard drives fail, however, is in non-sequential or random I/O. Hard drives can’t perform predictive analyses of data (though the computers they’re attached to conceivably can), so there’s no way to efficiently prioritize write events based on how future programs and users will access the data.

When you can no longer count on servers accessing data in an orderly manner — as is the case when you index files, edit video, record online transactions, simulate anything or run a virtual desktop infrastructure — the benefits of a hard drive fall short. Sure, you can store massive amounts of data, but if it takes two to three seconds for your servers to access it, it’s useless. VDI boot storms alone can make your hard disk array lock up and cause your users to revolt.

The Solution

Although hard disks aren’t optimized for random I/O, solid state drives are. There’s no front or back of a solid state memory chip, and accessing one sector takes the same time as accessing any other. It’s not a comparable solution for desktop PCs, because the cost per gigabyte of a bare solid state drive (SSD) is much higher than an HDD. For businesses, however, it’s a different story. A company can purchase an entire solid state array that employs deduping, thin/thin provisioning and compression to increase the virtual storage space far beyond the physical capabilities of each single array component. This eliminates the blender effect without compromising storage capabilities.

The world of enterprise data storage has evolved since the days of paper tape and punch cards; it’s changing more rapidly than ever now. If you want your servers to work for you rather than against you, switch from the old, slow hard disk to a better way.