Due to higher density and reliability, SATA drives are taking on more types of workloads, including those required for reliable and fastest backup. Some people remember the large hard drive study Google did in 2007 essentially showing no difference in failure of SATA versus more expensive SAS. Robin Harris did a good summary and links to the original article here. As for speed, RPM is one key metric for hard drives, as is access time(the time on average to position the heads above a given track and get to the needed sector). Many storage reviewers provide speed benchmarks with random or transactional (IOPS) performance of spinning drives, which is dominated by access time. Access time is, in turn, dominated by seek time, which is mainly the time it takes for a motor to move the read heads. However, in backup applications using imaging (which includes products from companies like StorageCraft, Symantec, VMWare , Microsoft, Computer Associates, Paragon, Double Take, Appasure, VRanger etc) data is read in relatively large blocks and written full tracks at a time sequentially. If the drive is empty the data is written to the outer tracks first, and then the read head is stepped in by one click and writes the next track. So access time is not a huge factor in backup. Backup performance is dominated by the RPM of the drive and the bits per cylinder (very dense hard drives like 2TB, 3TB, and 4TB drives have more bits per cylinder). Furthermore, as the drive fills up, the speed will decrease by as much as 50% as the heads move from the outermost to innermost cylinder. Since customers usually test backup speeds when the drives are empty, results may be skewed and backup take longer than they calculate as drives fill. As long as the interface speed is fast enough to keep up ( and 3Gps eSATA and 4.8 Gps USB3 interfaces arguably are), the interface speed has no measurable effect on sustained performance. The fastest drives today can sustain less than 200 MB/s (theoretically could backup data at 720Gigabytes per hour), which is less than the performance of a single 3GB SATA port.
What about using “multiple spindles” or RAID arrays to increase performance? When measuring IOPS many benchmarks show that multiple drives help. In fact, more drives equates to more speed. Our testing shows that with modern hardware RAID, our RAIDFrames and FirstRAID products can backup slightly faster than stand alone drives. But the difference isn’t night and day because again, for large sequential writes the speed advantages of multiple drives in small random writes disappears when write operations aren’t as dependent on seek times. Still, using our RAID 5 backup products is a good idea for redundancy of your important data. Plus you may gain 10 to 20% in backup and restore speeds with RAID. As for reliability, we believe slower spinning SATA drives have less heat and last longer than 10,000 and 15,000 RPM drives. This is true even of so called “enterprise” SAS drives. Studies of drive failures in large numbers by Google and others have proven that good old SATA drives have about the same failure rate as expensive enterprise drives. Because the density and sequential writes make up for RPM, they can be about as fast for backup applications too.