SCSI is going serial. Part 1 of this piece looked at the implications of SAS. We pick up the story here...
A less obvious benefit from SAS is that SCSI goes from being a parallel technology to a serial one. Current SCSI interfaces allow a maximum of 15 drives per cable, but the cables don't have the bandwidth to simultaneously support that many drives in real-world environments. The move to SAS increases the addressable number of connections per port -- without using expanders -- to 4032, compared with 127 for FC (Fibre Channel).
Because the connection to each drive is serial, there is no bandwidth sharing to prevent storage systems from using all those connections. SAS's serial nature also means that a failure of one drive cannot affect other drives. (The possibility of failure is one reason that SCSI has been such a highly reliable technology: The parallel architecture increased the impact of any one drive's failure.) In the past, the parallel approach was used because more data could be moved simultaneously per connection. But today's serial technology -- and the controllers to manage all the individual connections -- have progressed to the point where serial is the preferred approach, as exemplified by FireWire, USB, FC and Ethernet connections.
SAS also promises to break SCSI's performance barrier, with a current transfer rate of 3Gbps. "You can't drive parallel [technology] to any greater transfer rates. We're at the end of the bucket," Gartner's Monroe says. Initial SAS devices will also run at 300Mbps, and transfer rates should double to 6Gbps by 2008 and to 12Gbps by 2010, according to the SCSI Trade Association.
Click for larger view.SAS also adds support for dual ports, which provide two connections to one drive for fail-over reliability, as well as for redundant RAID controllers -- just as FC does.
The result of the greater device support and the faster transfer speeds will be larger clusters of drives in arrays with very high storage capacities and high performance. That should help SAS displace FC in all but the highest-performance storage tier, where FC's higher cost will be worth the better performance, IBM's Butler says.
SAS also has a use in near-line storage, according to LSI Logic's Mason. IT can aggregate several SAS devices -- arrays or external drives -- via direct SAS-to-SAS connections without having to put in a local SAN or share that network with other traffic, he says. Mason expects some enterprises to use SAS to create these near-line storage loops as an adjunct to, rather than replacement for, SANs. Similarly, IBM's Butler expects SAS to be used for two-node external storage clusters.
Still, vendors expect SAS to make its first mark in servers, which now rely on SCSI drives because of their high performance and reliability. SAS extends that performance and reliability while allowing vendors to use the same chip sets and connectors for their servers as they do for their PCs, reducing costs. The drives will become standard on servers by 2006, IBM's Butler notes. But because many enterprises have recently replaced their servers, SAS-based servers won't be deployed in large numbers until "the next refresh", Gartner's Monroe says.
Minidrives Two transition paths
Servers will also take advantage of the second SCSI trend: the move to smaller, enterprise-class drives. The 2.5-inch drives take less power and generate less heat, and their performance -- access and seek times -- is better than that of 3.5-inch drives because there is less distance for the drive heads to travel. That makes them perfect for transaction-intensive applications.
The small size provides another advantage. According to LSI Logic's Mason, 2.5-inch SAS drives mean blade servers will be capable of using reliable, high-performance SCSI technology rather than the 2.5-inch ATA drives -- designed for notebooks -- they've had to use in the past.
"We know we'll put 2.5-inch drives in a server or blade where there's less physical capacity," IBM's Butler agrees.
For drive arrays, Optimus' Hartzog expects midtier arrays to adopt 2.5-inch SAS drives first. Most midtier arrays hold 14 drives in a 19-inch rack, he notes, but could hold 30 drives if they switch to the 2.5-inch size. "You can add more spindles," Hartzog says.
"SAS will really capitalise on the move to 2.5-inch and outstrip the other technologies," LSI Logic's Mason says. "The small form factor is a fairly big deal."
Although 2.5-inch SAS drives help vendors deliver smaller drive arrays or increase the total capacity in the same cabinet space, Seagate's Castaldini says the 3.5-inch drive will remain in use for many years. That's because SATA drives don't have an enterprise-class 2.5-inch version, so to gain the benefit of mix-and-match enclosures, vendors will stick with the drives of the same size. Another reason is that the price of 2.5-inch SATA drives won't be as low as for 3.5-inch versions on a per-gigabyte basis for some time. "It's not clear what the cost per gigabyte is compared to 3.5-inch drives," IBM's Butler says, noting they will need to achieve price parity to gain broad adoption.
Nonetheless, change in the normally staid world of hard drives is in the air. Despite regular predictions of its demise, the hard drive has soldiered on, with storage density enhancements exceeding the transistor density increases postulated by Moore's Law. Soon the enterprise SCSI drive will take another leap forward -- this time in convenience, reliability, and lower cost of operation.