The most important disk drive facts are...
As drives shrink in size, the number of GigaBytes per cubic inch achievable increases exponentially.
As drives shrink in size, the number of Watts per Gigabyte consumed by the drives decreases exponentially.
The Mean Time Between Failures (MTBF) of a system is inversely proportional to the number of components in the system.
There are only two main classes of RAID.
Only one class of RAID can handle frequent failures without a catastrophic degradation in performance.
When using HDAs as disk components instead of complete drives, the sophisticated channel electronics - including the Low-Density Parity Check (LDPC) decoder - (which take a considerable amount of power) is not included in each component but is integrated along with other copies of the channel electronics in one large ASIC.
Disk database systems implemented using 1 inch drives or HDAs will require significantly less space and consume significantly less power than ones implemented using 2 inch drives or HDAs.
Having disk components fail fairly frequently in a large disk database system is not a problem. Have you ever heard of a department store having to close because too many light bulbs failed at the same time? With the proper amount of redundancy and regular maintenance, component failures are not a problem and will not cause any significant degradation in performance.
GigaBytes per Cubic Inch
As drive sizes decrease, the amount of data that can be stored in a cubic volume (GigaBytes per Cubic Inch) increases exponentially as illustrated below.
GigaBytes per Cubic Inch as a Function of Disk Diameter
A disk database system implemented using 2 inch drives will consume much more space than the same database implemented using 1 inch drives.
Watts per GigaByte
As drives are made smaller, the number of watts per unit of data (Watts per GigaByte) consumed by the drives decreases exponentially as illustrated below.
Watts per GigaByte as a Function of Disk Diameter
Most of the power consumed by an HDA is consumed by the motor in spinning the disks. Rotating disks are like a fan, there is a lot of aerodynamic drag, and the inertia of a disk varies as the fourth power of radius. In order to maintain the same degree of controllability, the power consumed by the motor also must increase as the fourth power of radius! Think what that means for disks of diameters 2 inch and 4 inch (14 = 1 and 24 = 16! ). In other words, a 4 inch HDA will probably consume more than 10x as much power as a 2 inch HDA and a 2 inch HDA will probably consume more than 10x as much power as a 1 inch HDA. This fact is extraordinarily significant!
Consider a disk database system implemented using 2 inch drives which consumes 1,000 watts of power. The same database implemented with 1 inch drives would probably consume less than 100 watts.
Two Main Classes of RAID
All RAID systems can be divided into two main classes - those where the drives in a drive group are written and read independently (RAID-4, RAID-5 and RAID-6) and those where all of the drives in a group are written and read simultaneously (RAID-1, RAID-2, RAID-3 and ECC Tek’s 2D-RS RAID).
Because of the write overhead penalty in RAID-5 type RAID, it simply is not practical to implement RAID-5 type systems if failures occur frequently. The write overhead inefficiency is too great and increases as the amount of redundancy is increased. The performance of large RAID-5 systems rapidly deteriorates catastrophically if failures occur frequently or if a large amount of redundancy is used.
Future Disk Databases
Future disk databases will use very small drives (as small as practical) because smaller drives consume much less space and much less power than larger ones. A sophisticated error correction system must be implemented because disks will fail fairly frequently. In order for the database to be reliable and always be available with high performance, frequent failures must be tolerated without any significant degradation in performance. Only the RAID-2 class of RAID (which includes ECC Tek’s 2D-RS RAID) can correct for frequent failures with no loss of performance, and the 2D-RS RAID scheme is the only RAID scheme needed because it can be configured to emulate every other RAID scheme.