Rachelle Trent, SAS Product Marketing Manager, PMC-Sierra
As enterprise storage requirements rapidly increase, expanding data throughput and providing scalable storage capacity is crucial. System Integrators (SIs) need to understand the new features of expander devices now, in order to be ready for market adoption of Serial Attached SCSI (SAS). The SAS standard specifies expanders, which are inexpensive switch devices connecting SAS hosts to SAS targets as shown in Figure 1 below.
Figure 1: A SAS Expander performs a data switching function between multiple hosts and target devices.
First generation SAS expanders, introduced in 2Q04, are widely available on the market today from several vendors. These expanders vary in the number of SAS ports provided, enclosure services supported and the size of topology or number of SAS devices permitted to be attached to the expander. In general, the end applications, whether server or external storage, dictate the feature set that an expander is required to support.
Next generation SAS expanders, available to the market later this year, will provide enhanced features to support more elaborate architectures for the external storage and blade server markets. These features include Topology Self-Discovery and Access Control Zoning and are detailed in the following sections.
Topology Self-Discovery
A SAS domain can consist of one or more expander devices. It uses tables within a subsystem. When a multiple expander device topology is included in a subsystem, routing of SAS data between expander devices is conducted via one of three routing mechanisms – direct routing, table routing, or subtractive routing.
An expander uses direct routing to forward connection requests from the SAS host to directly attached devices, it uses table routing to forward connection requests from the SAS host to other expander devices for further processing. It uses subtractive routing to forward unresolved connection requests when neither direct nor table routing succeeds.
Direct routing and subtractive routing are very simple processes and require no action from the SAS host. In the direct routing case, the expander requires no configuration from the SAS host because the routing applies to directly attached devices. For subtractive routing, an expander PHY must be initially set as the subtractive port, allowing any unresolved connection requests to be forwarded through the selected port.
Conversely, table routing requires the SAS host to dynamically configure the expander’s route table. The SAS 1.1 standard will provide two mechanisms for configuration of the expander route table. In the first method, an external management application residing in a SAS initiator (e.g. host bus adapter) configures the expander route table. In this method, the management application is required to traverse the entire SAS topology and discover the SAS device addresses within the SAS domain. The management application must then populate the route table for each table-routing expander with the SAS addresses of the devices within the domain. In the second method, the expander device acts as a self-configuring expander and is required to configure its own expander route table based upon its location within the topology of the SAS domain. In either methodology, SAS Management Protocol (SMP) commands are used to query and configure expander devices in the SAS domain.
First generation SAS expanders supported the first method of route table configuration – controlled by a SAS host. This method of centralized topology discovery, sourced from a management application residing in a host end device, is costly both in terms of the number of SMP transactions, as well as the number of inter-expander links consumed by these SMP transactions. In addition, SAS domains with multiple management applications/multiple hosts will incur large numbers of redundant SMP transactions as topology discovery operations are duplicated unnecessarily.
Second generation expanders will support the second method of route table configuration – controlled by the SAS expander itself. This method of a distributed topology self-discovery algorithm minimizes the number of SMP transactions and the number of expander links consumed by these SMP transactions. Furthermore, by configuring its own route table, the expander protects itself from any interoperability or configuration issues arising from one or more misbehaving management applications.
Access Control Zoning
SAS is gaining interest for use in small storage area networks (SAN). As a result, system integrators are driving for a feature addition to SAS to segregate traffic between devices. The requirement is similar to zoning provided in Fibre Channel and virtual LANs in Ethernet TCP/IP.
To respond to this demand, a zoning method, called Access Control, is being incorporated into the next version of the SAS standard. Second generation expanders will support this access control zoning. The zoning standard does not require changes to end devices; initiators will continue to perform normal SAS discovery and initiators and targets will perform OPENs normally. The access control zoning will simply limit the end devices such that they only see a portion of the SAS domain. A simplified way to understand the access control zoning divides, is visualizing that the expander is broken up into several sub-expanders or virtual expanders, as illustrated in Figure 2. The zoning of the expander limits a given host device to have permission to see only selected host and target devices. More information on the access control zoning standardization efforts can be found at www.t10.org. The access control proposal for T10 can be downloaded from ftp://ftp.t10.org/t10/document.05/05-144r0.pdf.
Figure 2: Access Control Zoning divides an expander into several smaller virtual expanders to restrict access between different end devices.
Summary
As SAS products come to market and the full system benefits of SAS are realized, system integrators are going to be looking for advanced expander features such as Topology Self-Discovery and Access Control Zoning. These two features will enable more complex topologies and ease the demands on the host devices in SAS systems.
