Quality of Service for SANs

SAN vendors are scrambling to satisfy customer demands for better management, security, and QoS features.

Over the past 10 years, developers have put enormous engineering effort into developing stable and highly reliable Storage Area Network (SAN) infrastructures. SAN development has emphasized resolving gigabit-transport technical challenges, host and storage SAN interface designs, and high-performance switch functionality. The underlying SAN infrastructure is clearly a prerequisite for supporting enhanced auxiliary services; however, the development of SAN management, security, and Quality of Service (QoS) features has trailed behind the deployment of storage networks. As a result, customers are now demanding better services to support their SANs, and the industry is scrambling to satisfy them. The Common Information Model (CIM) management initiative, for example, promises to provide a common framework that will let users manage SANs and storage resources from one console. Initiative groups and vendors must still address other services, such as security and QoS mechanisms for storage data.

QoS is an umbrella concept that includes acknowledging delivery and prioritizing different types of data for movement through the network (commonly referred to as Class of Service—CoS), guaranteed bandwidth through the network, and expedited data delivery from source to destination. Within a SAN, for example, online transaction processing (OLTP) data should take priority over less time-critical applications such as tape backup.

The QoS options available in current Fibre Channel SANs include Classes of Service that establish dedicated bandwidth and different levels of delivery acknowledgement. Class 1 service, for example, creates a dedicated link between source and destination, ensuring that no other traffic interrupts the conversation. Class 1 service supports acknowledgment, so a transaction's sending party can verify receipt. Class 2 service also supports acknowledgement, but doesn't require a dedicated connection through the fabric. Class 3 service is both unacknowledged and connectionless—and so provides no delivery guarantees. Nevertheless, Class 3 is the most commonly used Class of Service in Fibre Channel SANs. Why? Because Class 3 service imposes the least transaction overhead and therefore provides the highest performance. (Class 3 service, however, does rely on upper-layer protocols for recovery if frames are dropped because of congestion in the fabric.)

With Fibre Channel Classes of Service, you still can't identify and prioritize one type of storage traffic over another. No matter what priority users might place on their data, all data traversing the fabric is equal. Therefore, organizations often allocate a dedicated SAN island for each mission-critical and time-sensitive business application (and create separate SANs for backup and other less critical applications). Although this approach doesn't make the most efficient use of SAN technology, customers currently make this trade-off for lack of better QoS mechanisms.

Some promising approaches to QoS are evolving, however. For example, converting Fibre Channel-originated data into IP storage protocols lets SANs use QoS features originally devised for messaging traffic. Ethernet offers IEEE 802.1q traffic prioritization and Differentiated Services (DiffServ) options for flagging data streams with priority levels. For IP switches and routers, Multi-Protocol Label Switching (MPLS) or Resource Reservation Protocol (RSVP) can establish prioritized service and delivery guarantees through the network. Heterogeneous SANs with both Fibre Channel and IP components can thus supplement Fibre Channel Classes of Service and help users leverage large SANs that support a variety of applications concurrently.

Combining Fibre Channel and IP storage architectures can give customers additional tools for enforcing QoS in SANs. The ability to expedite delivery of mission-critical storage data, establish priority levels for different applications, and shape bandwidth to accommodate storage transactions lets customers maximize the return on their SAN investment and increases the end-user value of SAN technology.