SAN-based intelligence: the Holy Grail of storage management?

The buzz around "SAN-based intelligence" has sparked a lot of interest, according to recent polls of storage administrators. In spite of the dearth of real products, the major fabric infrastructure vendors and storage equipment vendors have been posturing themselves to capture mind share and the perception of leadership. But is there real value and viable solutions to SAN storage management headaches in this so-called network-based "intelligence"?

The SAN storage management puzzle today consists of many pieces including resource discovery, capacity planning, performance and availability tuning, provisioning, backup and recovery, disaster protection, and change management. Since data growth and application requirements are expanding so quickly, storage management can become a costly and painfully vicious cycle. Today's tools consist of host-based software such as volume managers, and array-based services like replication that don't work cooperatively together. Also, these services usually only work between homogeneous host-servers or storage systems locking users into a particular vendor with a specific roadmap. Keeping up with the plethora of management interfaces, software versions, and configurations only adds to the management complexity.

[FIGURE 1 OMITTED]

Centralizing and consolidating storage management services into a single layer that works across heterogeneous host-servers and storage systems is the proposed benefit of providing services in the network via "intelligent" purpose-built devices. The types of possible services include provisioning from a common pool of physical storage for better storage utilization, data replication between heterogeneous physical storage systems, performance information to tune configurations, remote asynchronous replication for distributing services like backup or for disaster protection, emerging advanced capabilities like dynamic data migration between different classes of storage (enabling Information Lifecycle Management), and changing the backup/recovery paradigm altogether with continuous backup and instant recovery services. Imagine the possible time and cost savings if managers could invoke all these capabilities from a single console and have them work with any servers or storage.

Sounds great, right? But how do we get there from here?

We won't get there very soon with so-called "intelligent" switches from the major vendors. Most have indicated that products won't be ready until 2004 or 2005. And even then, the capabilities are likely to be too basic to begin with. There are also products with fairly rich feature sets shipping today on so-called SAN appliances. However, since those solutions basically run on generic PC servers with HBAs, and reside in-band between application servers and storage systems, you'd better be sure you've got the scalability and reliability needed--not just for today, but for the next few years. That is, unless you're prepared to swap them out and start over again later. And, at the end of the day, a server with the processor horsepower, amount of memory, and quantity of HBAs needed for even modest-sized SANs can be quite costly.

One advantage of the appliance model is that it is not a fabric switch. The port personality of these appliances is that of a logical end-point like an HBA in a host server or a port on a storage array. That means you can add appliances to an existing SAN without disrupting the fabric. The downside of "intelligent" switches is that they are switches and are subject to the same interoperability concerns that exist today with fabric switches. What if you could deploy SAN-based storage services on an appliance-like platform that is fabric agnostic but has the performance, scalability, and reliability of fabric switches?

A new category of purpose-built appliances for hosting storage services in the SAN is coming on the market now--the Network Storage Services Platform. An NSS Platform is like an appliance in that the ports it provides are termination end-points, it is fabricagnostic, and can be installed non-disruptively in existing SANs. Unlike the appliance, though, the NSS Platform is built using custom hardware that provides a scalable number of ports with the performance and reliability of a fabric switch. An NSS Platform in the data-path between hosts and their storage should be just a "bump in the wire"--no different than a fabric switch.

The first NSS Platform-based storage service required for SAN-based intelligence is providing a target server at each port that logically terminates the I/O traffic like a virtual storage array head. Each target port of the NSS Platform can then be configured to export virtual disks (LUNs) to host servers using access control lists to ensure security. This "virtualization" layer has been hyped as the solution to the SAN storage management problem. But, in reality, "storage virtualization" is just the necessary enabling technology on which to layer services such as path management, volume management, replication, statistics monitoring, mirroring, and journaling.

Some NSS Platform implementations will provide integrated hardware and software running in a proprietary environment. While this type of implementation has the benefit of tight hardware and software integration (like fabric switches running fabric services software), the disadvantage is the inability to support a rich offering of storage services software from a variety of independent software vendors. You're pretty much stuck with the software roadmap from the vendor of this type of NSS Platform.

Another approach, albeit more difficult, is to make the NSS Platform an open platform running a commercial operating system, with storage services-specific APIs that allow ISVs to port various storage applications to the platform. It's also important that the level of abstraction provided by these APIs and the porting architecture be carefully designed to minimize the development and testing effort. If done right, applications can be ported and brought to market quickly at reasonable prices. The more difficult the porting job is, the more it limits the number and type of applications that can be brought to market, resulting in longer time-to-market with higher product costs.

[FIGURE 2 OMITTED]

The open version of an NSS Platform isn't just a static point product limited by a single vendor's ability to bring out new features. Rather, the open NSS Platform becomes a dynamic network footprint on which rich applications can be added over time to continually upgrade the centralized storage management capability of the SAN. This is a pragmatic, solutions-oriented approach to incrementally and non-disruptively adding intelligence to the SAN without having to "rip and replace" existing fabric, and without getting locked in to another proprietary software environment.

The open NSS Platform model has great promise, but what storage administrators really need are solutions to problems, not more complex technology to digest. And since many already have host-based software, existing SAN fabrics, various array-based storage services, and procedures and practices to manage these today, adding SAN-based intelligence should be an incremental and complementary project with measurable benefits. Fortunately, the NSS Platform can provide a rich toolset to cost effectively solve problems today without "boiling the ocean."

[FIGURE 3 OMITTED]

[FIGURE 4 OMITTED]

[FIGURE 5 OMITTED]

[FIGURE 6 OMITTED]

For example, if online backup of an application server is over-stretching its backup window, a network-based snapshot capability can solve the problem. By inserting an NSS Platform to re-host the application servers' storage volume, an instant snapshot capability can create a copy volume that can be backed up offline by a separate backup server. Instant snapshots and copies can also be used to do application testing against production data without disruption of production applications. And since the NSS Platform snapshot capability lives in the network above the storage systems, it's not necessary to provide free disk capacity from the same, homogeneous storage type as used for the primary volume. If the primary volume is provided on expensive, monolithic storage, the snapshot backing-store can use less expensive ATA-based storage for offline backup or application testing purposes.

Another example use of the NSS Platform is to bridge SAN islands, particularly those using different vendor fabric switches. Often, smaller SAN islands are built around applications like databases, e-mail, or file servers. This approach is simpler to design and administer and confines the business justification to the application being deployed. However, these islands may use different vendors' fabric switches and storage systems--making it difficult to bridge them for the benefits of consolidation. One benefit of broader consolidation might be to take advantage of a SAN-attached tape library for centralized backup. An NSS Platform can be added to bridge each of the SAN islands to the tape library SAN. However, since it is not a switch, the ports of the various servers and storage devices in each island don't actually see each other--meaning there is no complex zoning to manage. Instead, the bridging between the SAN islands occurs through virtual volumes. In the centralized backup case, snapshots are created by the NSS Platform for each volume in each island that needs backup. Those snapshot volumes can then be assigned to the backup server in the backup SAN. This approach is simpler than trying to conjoin heterogeneous (or even homogeneous) SAN fabrics, is non-disruptive to the applications running in each island, and is more secure since the NSS Platform acts like a storage firewall between SAN islands with only selected volumes exported between SANs.

Will storage management benefit from the "Holy Grail"--the elusive SAN-based intelligence being promised? SAN "intelligence" resides everywhere today, in hosts, in storage arrays, and soon in the network itself. Migrating storage services to the network should depend on getting a complete solution that is incremental, complimentary to existing host-based and array-based intelligence, and complementary to existing networks and preferably fabric-agnostic. If approached this way. SAN-based intelligence--using elements like NSS Platforms--will have a major and long-lived impact on improving storage management and costs.

Bill Terrell is CTO of Troika Networks (Westlake Village, CA)

www.troikanetworks.com