In part one of this feature we looked at the emergence of 2.5 inch drives. Now we examine the new IOPS/U concept and HP's plans for smaller drives.
Seagate envisages 1U RAID arrays and blade storage arrays. Comparing the new 1U servers to their 2U predecessors, the new 1U rack systems will offer 6 drive functionality (rather than 4) allowing RAID 5 and full mirroring. They draw 40 per cent less power while delivering 50+ per cent more IOPS-per-U of rack space and allow better airflow within the server. The company states, "The performance density or IOPS-per-U offered by these new systems will create storage solutions that are 33 per cent to 200 per cent smaller than today's systems."
This notion of IOPS/U seems novel. All drive manufacturers quote drive IOPS, but only Seagate is making use of the IOPS/U concept. Here is another example; "Today's storage arrays are moderate performance storage solutions. Ordinarily, 14 drives are placed in a 3U storage array or JBOD (just a bunch of discs) enclosure. The 14 slots use either 10K or 15K drives. With the advent of a smaller, high performance drive, approximately 30 drives can be bundled together in a 2U enclosure creating the fastest, highest performance storage arrays. The new 2U system will outperform today's common 3U rack storage array by nearly 140 per cent on an IOPS-per-U basis."
If it was a really important measure then we might consider using 1.8 inch drives to increase it further, which leaves this writer thinking that it might be a less than front rank consideration. It's also possible that the main IOPS restriction is the number of read/write heads a drive has. It seems intuitively obvious that a head can only be occupied with a single IO at a time. To increase the IOPS number you could increase the number of heads per drive or per platter. You could increase the rotational speed as well, shrinking the time per IO. Focusing on the drive's physical dimensions might just be a case of putting one's best marketing foot forward.
Another example of this is a comment from Ned Thanhouser, marketing director for Intel's enterprise products and services division, about Seagate's 2.5 inch drives: "Smaller drives draw less power and are easier to cool for rack-dense, highly reliable, server and storage solutions". That obviously depends on how closely you pack them together. The more there are in a particular space the greater the power needs and the harder the cooling job. What on earth is it to do with Intel anyway? The axe that firm is grinding is that its chips are used by the Seagate drives.
What both Fujitsu and Seagate are considering is that blade servers will need blade storage cards as well. We are looking forward to bladed offerings which offer complete server and storage farms in a rack. Only the 2.5 inch form factor can populate these arrays to a high enough capacity level and offer the IOPS levels needed.
Up until now Maxtor and Western Digital have made no public utterances on this topic. That can be expected to change in 2004. Seagate is telling us that products are due soon: "The (2.5 inch) drive platform will be produced for the current versions of Fibre Channel and Ultra320 SCSI, as well as Serial Attached SCSI. Systems based on the 2.5-inch enterprise platform will appear in 2004."
Hewlett Packard is supportive of the 2.5 inch enterprise drive idea. Jeff Jenkins, HPs acting VP for industry standard server storage and infrastructure, said, "In a consolidated IT environment, our customers need to provide more I/O's in the same rack space while being able to scale to higher I/O's and capacity to meet future growth requirements without increasing their data center space. The density-optimized HP ProLiant DL server line and the direct attached storage systems can be optimized with 2.5-inch enterprise-class disc drives using future interfaces such as Serial Attached SCSI to enable our customers to receive performance enhancements without the expense of adding more racks or data-center floor space."
If Seagate and Fujitsu are successful and other manufacturers bring 2.5 inch drives to market then we can expect them to spread into the desktop and non-enterprise server markets. But what about the lower capacity of these smaller drives? "To step down to small form factor means dropping down to the next capacity point, say from 300GB to 146GB," according to HP's Jenkins. But HP's most popular drive is a 36GB model. The new Fujitsu drives go up to 73.5GB and more, much more can be expected, so the point is moot.
Another effect may be that entry-level servers and desktop PC cabinets may become smaller. Multi-disk desktops may also become more usual, leading to more fun and games with Windows, boot partitions, backup and a new Windows file system.
Lastly, we might wonder what effect having more disk drives will have on backup. Simply having greater disk capacity per system will tend to increase backup time and increase the need for multi-reel backup sets. Both points could speed the introduction of disk-to-disk backup as an end in itself or as a staging post on the way to an ultimate tape backup.