Author: Charlie Kraus, Director, HBA Business Unit
Adoption of the next evolutionary step in SCSI technology, Serial Attached SCSI (SAS), is well underway, with announcements from major OEM storage vendors starting to appear. Initially the role of SAS was understood to be the next direct-attach technology for internal disk drives in servers, or for external drives in SAS JBODs. We will see just such applications of SAS technology in the beginning. But there are other aspects to SAS that will have wide-ranging implications and prove disruptive to current strategies for building storage systems.
The first such aspect is the compatibility with Serial ATA (SATA) technology. Controllers for SAS also support SATA drives, allowing a mix of drive types in a common enclosure. This mixing capability enables the deployment of different drive types for different storage requirements – high-performance SAS for near-line fast access storage, and low-cost SATA capacity for long-term storage. While these types of systems are possible with Fibre Channel (FC) and SATA drives, each requires a separate controller, adding cost and diminishing flexibility. SAS/SATA mixed configurations allow dynamic changes in the mix of drive types.
The more important aspect of SAS/SATA technology, as far as disrupting current thinking on storage system design, concerns SAS expanders. The basic use of expanders will be to allow a controller to support many more drives than just a single drive per port. Twelve-, 24- and 36-port expanders are appearing that will be typically deployed in SAS/SATA JBODs. The SAS controller in a server will thus be able to interface to many drives in the JBOD:
Figure 1: A 36-Port Expander is Connected to 32 Disk Drives, with Four Ports Available For external Connections.
Even more important than expanding the SAS controller ports, expanders essentially function as switches. This enables expanders to play a more complex role; instead of a single server interfacing to a single JBOD, multiple servers can be connected to multiple JBODs as shown in Figure 2.
Figure 2: Example of Blade Servers, SAS Expanders and
Drives in a Cluster Configuration
SAS and Blade Servers
This combination yields very powerful capabilities and can be thought of as a SAS fabric. But we have to be careful not to confuse this fabric capability with FC or Ethernet fabrics. SAS has an important difference with FC and Ethernet: SAS traffic is not connectionless switched-packet based like FC and Ethernet, but is connection-oriented. SAS also does not have a transport layer. SAS expanders allow switched serial interfaces.
Consider the implications for scalability: SAS fabrics will likely be relatively small, less than 100 ports, as early SAS expander chips will offer 12, 24, 28, or 36 ports, limiting how many can be cascaded to build larger port-count switches. In the future, higher port-count expanders will allow larger switches.
For Figure 2 above, around 72 ports will suffice. What will make this configuration especially compelling is the much lower cost of a SAS fabric and its ability to support both SAS and SATA. SAS fabrics will likely be one fourth of the price per port (or less) than competitive solutions. SAS also allows diskless servers to boot from a SAS disk – very easy to do across a SAS link – thereby preserving the legacy SCSI boot model and not requiring boot-from-SAN capabilities. SAS switches can even support storage applications such as virtualization, replication and RAID.
For larger SAS storage infrastructures, there will be a need to connect multiple SAS switches via a longer link than the eight meters that SAS copper cables allow. The first thought may be, “Why not support SAS over a fiber link?” The issue is the control plane signaling, which does not run at the 3Gb/s frequency of the data. With copper it is not an issue to have multiple frequency signals, but fiber transceivers running at the 3Gb/s data rate will not carry the control plane signals at another frequency. What are the options? Both FC and iSCSI can go long distances. The choice will depend on what is currently installed in the data storage center.
Given that SAS/SATA will appeal to companies that do not have FC installed, the most popular choice is likely to be iSCSI. SAS switches could have an optional slot to plug in an iSCSI port to allow expansion across long distances, in fact anywhere on the Internet. But do not forget that SAS device discovery is defined within a SAS domain today. To accomplish discovery across an iSCSI link will require development work; not difficult, but necessary. The connection-oriented SAS traffic will have to be converted to iSCSI packets at the iSCSI interface on the transmit side and reversed on the receive side.
In summary, we can expect that SAS/SATA technology will have a dramatic impact on the way storage infrastructure is configured, especially for mid-range (and below) companies. New vendors will likely gain power, providing solutions in this space against entrenched FC players. The next two years promise to be exceedingly exciting.