Intelligent SANs: issues to consider when selecting an enterprise-class network storage controller.

Despite the fact that storage costs continue to drop, most data centers still find it too costly to store all enterprise data on a single platform. Yet the alternative--selecting a best-of-breed storage solution for each application--while certainly cost-effective, leaves data centers with a disparate environment where different vendors' SANs, storage management software, and media types exist as isolated islands.

To resolve the complexity of managing heterogeneous storage architectures, the concept of delivering intelligence within the SAN was developed. This "intelligent SAN" presents a storage abstraction layer to servers within the environment, enabling integration of disparate resources into a virtualized pool that is available horizontally across the entire data center and all applications When selecting an intelligent storage network solution, data center managers face a growing array of choices, each promising to deliver the virtualization and storage management capabilities they require.

Homogenous Management of Heterogeneous Resources

Consider at the outset that although intelligent storage networks were originally conceived of as a way to simply virtualize disparate resources, they can also offer a number of value-added functionalities, the most notable being data mirroring and replication across heterogeneous systems. In addition, some solutions can also facilitate load balancing across storage media, and support implementation of a quality of service hierarchy that enables cost effective service level agreements without the need to over-provision storage assets or network bandwidth.

[FIGURE 1 OMITTED]

But before evaluating what specific features are required or desirable, remember that meeting the fundamental objective of consolidating heterogeneous resources in a data center environment requires that the infrastructure can scale as data center storage volumes grow--without performance bottlenecks or single points of failure. Unfortunately, however, not all solutions meet these fundamental requirements.

External server-based appliances that are implemented within a storage network, for example, actually create a bottleneck that can negatively impact application performance. When this occurs, additional appliances have to be added--each becoming a separate management point in a costly move that, at best, can only solve the overload problem temporarily. In other words, as storage volumes increase and new resources are implemented, additional appliances will have to be implemented as well, adding to management complexity and overhead. Additionally, these appliances also create a single point of failure within the data path which, in turn, affects application availability.

To address the availability issue, some server appliances are implemented "out-of-band", providing virtualization and management capabilities out of the primary data path. But users of such solutions are still left with a never-ending cycle of adding resources that defeats the key benefits of an intelligent storage network--management simplicity and reduced storage costs in a highly available, highly scalable architecture.

In an attempt to rectify the shortcomings of external server-based appliances, some SAN switch vendors have introduced in-band application modules within the switch chassis--an approach that eliminates the need for an external device. Unfortunately, though, this approach does not solve the performance, scalability or availability problems. The reason the scalability problem remains is that virtualization and management functions are implemented on individual application modules (server blades) within the switch, each with a limited capacity, yet each having to serve a potentially large number of ports. These application modules also create a significant cost barrier, costing as much as $200,000 each, supporting as little as two terabytes of data. Furthermore, in some cases only a very limited number of modules can be supported per chassis. As a result, users have large upfront costs and may run out of chassis slots (and ports) before the bottleneck problem can be resolved.

[FIGURE 2 OMITTED]

Of equal concern is the fact that, as with external appliances, each blade can only manage and virtualize storage resources attached directly to it. Or, put another way, appliance and switch-based intelligent storage networks both suffer from discrete capacity limitations that prevent enterprise-wide resource sharing, do not lower storage resource costs, and do not optimize data lifecycle management across the entire data center.

Delivering on the Promise of Intelligent SANs

The only way to completely resolve the capacity limitations of appliance and switch-based intelligent storage networks without continually facing the need for adding servers or blades is with a solution that offers in-band, distributed, port-based intelligence rather than centralized, server-based intelligence. This paradigm, often referred to as an enterprise-class network storage controller, delivers instant scalability as a data center expands and adds storage resources. Storage controllers also allow storage resources to be consolidated and virtualized with a single homogeneous management interface that can uniformly allocate, replicate and mirror data across heterogeneous storage resources throughout the enterprise.

Tiered Storage TCO Reduction

This inherent scalability across the entire data center on a port-by-port basis carries immediate bottom-line benefits because it enables data centers to build a tiered storage architecture, managing low-cost, mid-range and high-end storage arrays across a single infrastructure. In a typical 50-terabyte data center where just half of all storage capacity is in the form of modular RAID arrays (costing about $0.03 per megabyte) instead of monolithic arrays (costing about $0.30 per megabyte), first year costs can be reduced nearly 50%, from $15 million to $8.25 million. By introducing serial ATA (SATA) arrays (which can be as inexpensive as $0.005 per megabyte), cost savings can be even more dramatic.

The savings are equally impressive when expressed in terms of cost per VIOPS (Virtual I/O Operations Per Second). For a Maranti enterprise-class network storage controller, the cost per VIOPS--a measure of how fast a pool of virtualized data can flow from a storage array to an application server--is approximately $0.06 per VIOPS. By comparison, the exact configuration for a leading appliance-based SAN switch solution can cost as much as $2.16 per VIOPS (36 times higher). The reason for this enormous difference is the fact that where the SAN switch solution can only support 280,000 VIOPS across two blades on a single chassis, a Maranti network storage controller can support over 6.4 million VIOPS within a chassis.

[FIGURE 3 OMITTED]

[FIGURE 4 OMITTED]

Another benefit of these enterprise-class network storage controllers is that they are purpose-built hardware and software to provide wire-speed virtualization, replication, and mirroring across heterogeneous storage resources. SAN-based intelligent storage networks, on the other hand, are often cobbled together from different vendors' products with a software port from one or many Independent Software Vendors (ISVs). As a result, scalability, integration, and maintenance can be problematic, and if an issue should arise, there could be some difficulty assigning responsibility. With the Maranti storage controller, integration and scalability are never issues, and there is sole source responsibility, offering the end-user a "single throat to choke," in the unlikely event of a problem.

Selection Criteria

Beyond the ability to virtualize storage, which is really the defining attribute of any intelligent storage network, consider whether the solution can also replicate and mirror data at each port across a heterogeneous, virtualized storage resource. Be aware that when scalability is tied to adding servers or blades, this capability is unlikely to be available. But with Maranti's storage controller, not only can data be replicated and mirrored across the entire data center storage pool, but the low latency, wire-speed performance of these operations is always maintained as solutions are scaled to increasing numbers of ports and arrays.

One way to eliminate any performance degradation is to ensure that virtualization, replication, and mirroring can all be executed at line-rate (2Gbps). Again, this is unlikely to be the case when solutions require multiple servers or blades. On the other hand, with distributed network storage controllers, any port, anywhere in the data center can have wire speed access to the data center's entire pool of virtualized storage. This capability enables data center managers to allocate resources where and when they are required.

One key benefit of port-based resource allocation is that it enables quality of service (QoS) policies that let data center managers differentiate data flows to ensure that mission critical applications always have priority access to storage resources. Intelligent storage networks that support QoS functionality enable utility-based service level agreements (SLAs) without the need for over-provisioning.

With Maranti's network storage controller-based QoS, for example, straightforward policies can be established that simply assign storage resources on the basis of an application's profile, thereby eliminating the high costs of over-provisioning. By comparison, without QoS, SLAs can only be met using brute force; implementing 300% more capacity than is required during day-to-day operations just to ensure storage resources are available during peak periods is not unusual.

To optimize performance for key applications, Maranti network storage controllers also support load-balancing features that can significantly boost read I/O performance by spreading reads across multiple physical storage arrays, creating mirrored copies of data for availability, and distributing I/O processing for an application to improve performance.

Performance can also be enhanced with an intelligent storage network that automatically classifies storage into multiple tiers on the basis of pre-determined policies. In this way, each time a new disk is added it can be automatically discovered and assigned to the appropriate pool of resources without having to reboot application servers or reallocate resources. As a result, storage resources can be instantly mapped to information lifecycle management policies, and each application can be assigned to an appropriate storage pool.

Not to be overlooked is the importance of high availability. Simply put, if an intelligent storage network solution is not fully redundant and has a single point of failure, look elsewhere. Only with automatic fail-over to redundant chassis' can data centers be sure that their storage resources will be available 7X24XForever.

What it comes down to is this: An abstraction layer in a storage network can bring order to chaos, offering homogenous management control over heterogeneous resources. But the full operational and cost benefits of an intelligent storage network will only be available if the solution is not server or blade dependent, can virtualize unlimited amounts of storage, can replicate and mirror data across a virtualized data pool, and can prioritize storage access through quality of service features. The latest generation of enterprise class network storage controllers from Maranti delivers all of these benefits--and more. A careful evaluation of competing alternative clearly demonstrates that they simply cannot meet this standard.

Rick Walsworth is vice president of marketing for Maranti Networks (San Jose, CA)

www.maranti.com