Heterogeneous SANs: the "Circe" of storage.
For the IT organizations that fall entranced by this call, many often find that the reality of heterogeneous SANs is not so beautiful--making heterogeneous SANs the Circe of the storage industry. Circe was a legendary enchantress in Greek mythology whose charms few could resist. She enticed passing sailors to her abode only to transform them into beasts. Like Circe, heterogeneous SANs often change storage administrators into beasts of burden.
Problems with Heterogeneous SANs
Direct-attached storage (DAS) couldn't be simpler. Attach a disk array to a server, configure the LUNs, create the file system, and you are on your way. The SCSI protocol is elegant in its simplicity. That's because SCSI was designed just for this environment and not for a switched environment, such as a storage area network.
SCSI's introduction into a SAN environment creates a level of complexity because it places a point-to-point protocol designed for dedicated SCSI cables into a shared transport environment. Even for homogeneous SAN environments, this requires additional SAN security through LUN masking or zoning. As seen in Figure 1. storage administrators need to carefully map out the connections between the servers (and their respective host bus adapters, or HBAs) and the storage. Improper SAN security can lead to data loss or rogue servers gaining access to confidential data.
This problem escalates geometrically (n X m, where n is the number of servers and m is the number of storage devices) as you add more servers and storage to a single SAN. And this equation only takes into account the number of physical devices, when it's really the number of paths between HBAs in the hosts and volumes available from the storage that drive complexity. Moreover, there may be multiple paths to the same volumes to the same host or to different hosts in the case of clustered environments. Layering read-write access rights onto each path, along with the various implementations of dynamic multi-pathing software and high availability implementations within storage controllers, further complicate the environment.
Now introduce heterogeneous storage elements into the environment, different hosts with different operating systems and HBAs, different Fibre Channel switches and different storage devices. Suddenly, a complex but conceivably manageable environment becomes a management nightmare. The beast of burden transformation begins in earnest.
[FIGURE 1 OMITTED]
First, the easier part. Each of the storage elements in the infrastructure has its own element manager. While not ideal, the storage administrator will need to learn how to manage the various elements in the infrastructure via each individual element manager. Fortunately, device management complexity only scales linearly, based on the new devices introduced into the network. Moreover, with management frameworks, this can be handled satisfactorily.
Now, the more difficult part. Interoperability becomes nearly impossible to achieve. As new devices are added to the network, they add complexity into the interoperability matrix, Multipath management software may work with some storage devices but not with others. Similarly, host bus adapters on certain operating systems with certain firmware will work with some storage devices but not with others. Moreover, HBA firmware on hosts may be inadvertently upgraded by systems administrators without storage administrators' knowledge, leading to problems in the heterogeneous SAN. Detecting events in the environment and correlating them to a root cause becomes increasingly complex.
Now, the most difficult part. The storage management tools for these environments are quite immature. Often sitting out-of-band, they cannot provide both a centralized management view and the fine-grain virtualization that is needed to simplify the environment. Without the centralized management view, the complexity cannot be understood; yet without the fine-grain virtualization the complexity cannot be reduced. Consequently, the environment's operational costs are driven higher and expensive professional services knowledge is needed to create and maintain the environment. For those storage administrators entranced by Circe's call, the transformation to beast of burden is complete.
[FIGURE 2 OMITTED]
Moving Intelligence Into the Network
What is the answer? For most, they resist the call and remain with direct-attached storage or multiple homogeneous SAN islands, Unfortunately, they do not achieve any of the benefits outlined above. Their storage infrastructure is locked in a time capsule that can't evolve as IT business requirements change. This is not sustainable over the long term.
Stepping back and looking at the source of the complexity provides the answer. Since SCSI was designed to be a point-to-point protocol, by placing the intelligence into the network that connects servers and storage, one can reintroduce this simplicity back into the storage architecture. This intelligent intermediate element is called a network storage controller.
As seen in Figure 2, as far as the servers are concerned, the network storage controller appears as the storage. As far as the storage is concerned, the network storage controller appears as the servers. The environment is simplified as the storage controller functionality is brought to the key leverage point in the architecture.
Like Odysseus in Homer's Odyssey, who saves his men from a life as beasts, a network storage controller can relieve a storage administrator's tasks as a beast of burden. There are four elements of a network storage controller that are necessary for this success;
* Purpose-built hardware
* Fine-grain virtualization
* Centralized management
* Superior interoperability
With its position in the center of the data path, the network storage controller must be purpose-built in order to have the levels of performance and reliability, availability and serviceability (RAS) that are typically found in monolithic storage devices.
Consequently, the capability of providing active/active high availability within the SAN in conjunction with multipathing software is imperative. Moreover, sequencing the I/Os will allow the network storage controller to scale in performance as differing loads are applied to the system. Additionally, error handling and recovery needs to occur at the port level to isolate the errors from affecting other I/Os in flight. Advanced features, like tracing activities down to individual volumes and ports, are crucial for event management and correlation.
Fine-grain virtualization allows a network storage controller to mask the complexity of the servers from the storage and vice versa, Communication occurs directly with the network storage controller, which automatically handles delivering the right I/Os to the right location. There's no need for the server to understand the underlying storage infrastructure; space can be made available in seconds regardless of the storage type. In fact, individual volumes can be sliced and concatenated from multiple different heterogeneous storage devices, if desired. This approach also delinks the installation of storage from the provisioning of storage, allowing for all storage to be installed and made accessible to the network storage controller. At a later date, through single-step provisioning, it can be allocated to whatever server needs the capacity.
Centralized management provides a single dashboard view of the storage in the infrastructure. It can show the pool of storage under the network storage controller and, using profiles, create policies for provisioning storage to hosts. Different profiles corresponding to the underlying capabilities of the storage pool can be used to effectively implement tiered storage architectures. Also, since the network storage controller sits in a strategic location in the network, it can identify problems with the data path and identify error conditions, as well as conduct root cause analysis for event management and correlation.
Lastly, superior interoperability testing solves the "n x m" complexity problem. By ensuring interoperability between the network storage controller and the various elements within the infrastructure, interoperability within the heterogeneous SAN is ensured. Moreover, interoperability can be achieved without reducing everything to its least common denominator. Rich device characterization allows for a variety of benefits including storage profiles, event management and correlation and tiered storage.
Reducing Cost, Complexity and Risk
Indeed, the call of heterogeneous SANs can be achieved. By recapturing trapped storage, storage administrators can increase storage utilization by more than 30% without purchasing new hardware, saving real dollars. Moreover, vendor lock-in can be avoided leading to increased competition, and tiered storage can be implemented, reducing storage acquisition costs by at least an additional 25%.
Increases in operational costs can be minimized with a policy-based storage management approach and by improving the efficiency of existing storage administrators. Expensive professional service engagements can also be avoided.
Lastly, with heterogeneous SANs, an IT organization can deploy a storage infrastructure that helps drive increases in revenue by quickly deploying timely storage for peak loads or new business requirements. Also, with more data online through tiered storage, the information can be better leveraged for competitive advantage.
Must heterogeneous storage be the "Circe" of storage? With new approaches to delivering intelligence into the network, the answer is no. According to Homer. Odysseus and Circe eventually lived happily together for more than a year--which was one of the best times during his travels. With intelligent network storage controllers, storage administrators can see the same happy ending.
Stephen Terlizzi is vice president of marketing and business development at Candera, Inc. (Milpitas. CA)
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||Storage Networking; Storage area networks|
|Publication:||Computer Technology Review|
|Date:||Feb 1, 2004|
|Previous Article:||Maintaining quality of service for WAN storage over IP.|
|Next Article:||Serial attached SCSI and serial ATA seek their levels.|