The next evolution in storage: clustered storage architectures - Storage Management

Computer Technology Review, Sept, 2003 by Rob Peglar

In the no-excuses 24x7 business environment the IT infrastructure must adapt dynamically to unpredictable utilization and processing requirements, and must deliver uninterrupted uptime. Clustered processing and application architectures have been proven to improve uptime and resilience, configuration flexibility, and make it possible to increase capacity as needed.

In contrast, in the enterprise storage arena, storage area networks (SANs) continue to feature dual-processor, chassis-based, static architectures--limiting storage flexibility, increasing complexity, and requiring significant up-front capital investment. To protect against unplanned downtime events chassis-bound architectures rely on a complex system of proprietary backplanes, interconnects, and processors. Organizations have little control over these proprietary infrastructures, severely limiting configuration flexibility.

Moreover, legacy architectures demand planned downtime when configuration changes or upgrades are required to match ever-changing business needs. Although planned downtime accounts for over 80% of system unavailability, the industry accepts this as "status quo." In today's Web-based business environment, the impact of any type of downtime--planned or unplanned reaches far beyond internal users: it can impact revenue generation by impacting customers, just-in-time (JIT) supplier delivery and partner performance. Planned downtime limits overall responsiveness, increases operating expense and total cost of ownership (TCO).

Although clustered storage is a departure from legacy SAN architectures, computing and application administrators already understand the nature and management of clustered solutions. By applying this expertise to the storage arena, organizations gain the same proven benefits of application and server clustering in the storage infrastructure.

Clustered storage delivers resilient, continuously available information access to essentially eliminate both planned and unplanned downtime. A distributed, clustered storage architecture changes the way storage is deployed and managed across the enterprise. A modular design enables on-demand scaling of capacity, performance and availability, matching storage capacity to ongoing business requirements and fully utilizing storage resources.

The resulting savings and efficiencies, both in operational costs and capital investment, dramatically improve the bottom line. The clustered approach provides a storage architecture that delivers true competitive advantage, business responsiveness, and accelerated return on investment.

A clustered storage architecture offers improvements over a traditional SAN architecture because it supports:

Multiple layers of resilience: Traditional architectures focus on eliminating single points of failure within a chassis-restricted system. Storage clusters can deliver Multiple Layers of Resilience (MLOR). MLOR enables a storage architecture to provide resilient availability, performance and data integrity across infrastructure layers--in many cases where traditional architectures would suffer from performance degradation, data loss or outright failure. Storage Clusters improve resilience beyond traditional single-chassis systems by supporting distributed, N-Way availability, continuous storage performance, full server and application infrastructure independence and maximum data integrity in case of site or cluster-wide disasters. This approach to the architecture delivers resilience far beyond the traditional single-chassis focus on elimination of unplanned downtime, delivering distributed availability.

Fiber-distributed storage: The clustered storage architecture leverages modular components, deployed independently across fiber-distributed networks. When components are placed where they are needed versus in a single chassis, flexibility and resilience increase to mitigate risk in the event of an outage in any single location. This architecture can give IT organizations the flexibility to design and deploy a storage configuration that matches their specific reliability and cost requirements. A relatively simple configuration might involve a single interconnect path between two nodes, whereas a full N-way mesh configuration will provide ultimate resiliency.

N-Way clustering: With N-Way storage clustering, multiple storage processors (controller nodes) can be deployed across fiber networks, and controllers can be added as needed to provide greater resiliency and performance. If controllers provide an active-active I/O workload and communicate between all controllers within the cluster, fiber-distributed controllers will be able to continue to process storage requests even if all but one location has all outage. Servers and applications will continue to access, process, and deliver business information--transparent to failure events.

Multi-level failover/failback: Multiple layers of resilience within and across the storage cluster can be provided by failover and failback mechanisms that optimize storage resilience and continuous application performance. The following describes the levels across which failover and failback mechanisms might operate and the advantages of each.