Disk and Array Selection Criteria

Today, arguments for and against platforms based on "high-end" (UltraSCSI, SAS, FC and iSCSI) and "low-end" (UltraATA and SATA) drives and interfaces are appearing in the trade press and on conference agendas that are equal parts marketing hype and techno-babble. The situation is creating confusion for many planners as they seek to establish an intelligent foundation for their storage infrastructure.

In the selection of disk and interface technologies for your next array purchase, the following criteria will help sift through the marketecture:

1. The application is king. To select an appropriate technology for storing application data, the nature of the data and its characteristic access, security, and protection requirements must be identified and understood . Table 6-2 describes some typical characteristics of modern applications.

   Table 6-2. Typical characteristics of application data

   Application	Read Intensive	Write Intensive	I/O Intensive	Throughput Intensive	Random Access	Sequential Access
   Online Transaction Processing	X	X	X		X
   Data Warehousing	X		X		X
   File Services	X			X	X
   Medical Imaging		X		X	X
   Web Services	X			X	X
   Multimedia Video	X			X		X
   Document Imaging		X		X		X
   CAD and GIS	X		X		X
   Backup/Restore		X		X		X
   Archive		X		X		X

2. Data protection is paramount. The data that is being stored must participate in some sort of data protection strategy. For mission-critical data, most organizations seek to architect some sort of real-time redundancy into storage associated with the data. This will usually require dual-ported drives that can be attached to two, rather than one, controllers in an array as a protection against single point of failure (see Figure 6-9). Higher-end drives are dual-ported: typically, lower-end drives are not. However, an important caveat raised by many ATA array builders is that fault-tolerant arrays can be readily constructed using double the number of single-port drives and specialized controllers that still cost less than the total price of an array comprised of dual-port drives (see Figure 6-10).

   Figure 6-9. Dual-ported drives for high availability. ^[8]

   

   Figure 6-10. Single-ported drives in high availability array configuration using routing/switching/multiplexing (RSM) technology. ^[9]

   

3. Data integrity is key. For critical application data storage, protection is required to ensure that data written is data read. Top vendors have adopted "long blocks" in their high-end drives: sector information is supplemented with an appendix that does block checking to validate the address and contents of the sector (see Figure 6-11). This is an important stopgap against firmware and hardware errors.

   Figure 6-11. Long blocks for data integrity.

   

4. Mean Time Before Failure (MTBF) must be considered . High-end drives are designed to provide a service life of more than 1 million hours before failing. Engineers design the drives for 24-hour/7-day operation (8760 hours/year). The drives in personal computers are designed for an 8- hour /5-day (2080 hours/year) duty cycle and typically demonstrate greater vibration as a function of their design, contributing to an abbreviated MTBF as well as greater frequency of write aborts and read retries. ^[10]

5. Performance requirements must be considered. Vibration can also impact speed of drive operation as seek times are protracted. That is part of the explanation for why high-end drives tend to provide 40 percent better performance than lower-end units. The other part is a combination of features of high-end drives engineered into units specifically to up performance. Better motors, more precise actuators, better electronics, and enhanced firmware providing such features as sorted work queues all contribute to much improved drive performance at the high end.

6. Warranty provisions may also be important. Put bluntly, high-end drives tend to have better warranties ”testimony to the vendor's belief that longer MTBF will minimize warranty replacement expense.

From the criteria above, it may seem that high-end drives are the preferred foundation for any storage infrastructure. They might well be if cost were not an issue. In the real world, however, budgets do not permit the universal hosting of all data on the most expensive platforms.

Cost efficiency is gained by first matching a drive and interface to application data characteristics, then using practical parameters such as budgetary constraints to cull the list of possibilities. For virtually any application, there is an optimal storage solution and a number of less optimal ones from the standpoint of pure technology that may be better suited from a business value standpoint.

In the current industry view, disk and interface technologies may be viewed as forming a matrix bounded by performance and capacity characteristics (which are inversely proportional to one another) on one side and SCSI or ATA on the other ”rather like the one depicted in Figure 6-12. ^[11]

Figure 6-12. Industry matrix model of disk/interface solutions.

Another approach, most often championed by vendors of SAS and SATA, is to base disk and interface selection on the frequency of data change. If data is relatively nonchanging (read more often than written or updated), lower-performance, high-capacity disk may be preferred. ^[12] Conversely, for data that changes frequently, high-performance disk may be the better solution. Simplistic as this division might seem, it has the advantage of clarity without engaging in hyperbolistic infighting over one interface protocol or another.