Following sound SAN design strategies allows you to reduce the number of SAN islands, strengthen your primary SAN,...
make storage easier to manage and provide more data protection.
Project-based and departmental SAN islands are multiplying faster than rabbits and are relatively inexpensive, easy to build and require little planning. But SAN islands carry a hidden cost: The benefits of isolated SANs vanish rapidly as more of them are added to the storage infrastructure.
A SAN island's single-purpose nature makes it very uneconomical, and the aggregate cost for storage and storage management of small SAN islands can be eye-opening. This wasn't an issue during the dot-com years when funds seemed inexhaustible, but the ensuing economic dip ushered in an era of consolidation and centralization; in other words, doing more with less. SAN consolidation has the following benefits:
- Simpler storage management
- A higher degree of storage resource utilization (capacity, bandwidth, storage controller, I/Os, storage management staff)
- Increased reliability of the storage infrastructure
- Enables tiered storage
- Simplified compliance and disaster recovery (DR) processes
The big picture
Any consolidation project needs to start with an inventory of existing storage arrays, switches, servers and applications. A good understanding of the existing storage landscape, applications and user requirements is needed before you can assemble a plan to move toward a more consolidated SAN.
|SAN consolidation tips|
Redundancy and the required level of SAN isolation need to be a central part of the consolidation plan.
A vendor review, with an eye toward possible vendor consolidation, should be included in the storage consolidation project.
If a storage chargeback model is in place, establish internal prices for the different storage tiers.
Make sure the proper storage management tools are in place to monitor SAN performance and availability.
Obviously, the importance of planning increases with the size of the primary SAN and the number of smaller SANs that need to be consolidated. While little planning is required when building small, isolated SANs, thorough planning is imperative when there are multiple departments and applications competing for consolidated SAN resources. The "buy storage" mindset that dominates isolated SAN deployments needs to make way for an "architect storage" mentality. Large, consolidated SANs aren't bought--they're designed and architected.
The SAN consolidation project must address the two biggest concerns of tying SAN islands into a data center's primary SAN: an increased level of dependency on the primary SAN's availability, and possible conflicts among applications and users accessing the SAN. Both concerns can be addressed with an appropriate SAN design.
The risk of downtime is reduced by putting the proper level of redundancy in place. Some of the cost savings realized by consolidating storage needs to be reinvested in a redundant storage design that eliminates single points of failure or at least substantially reduces the possibility of downtime. "The more eggs you put into one basket, the more robust you want your basket to be," says Mario Blandini, director of product marketing at Brocade Communications Systems Inc.
Application and user conflict concerns can be addressed with SAN isolation techniques like zoning, LUN masking and Fibre Channel (FC) routing with prioritization methodologies like quality of service (QoS) and bandwidth reservation offered by contemporary SAN equipment. The key to overcoming conflict concerns lies in a proper SAN architecture that guarantees the right level of performance for each application. A disk-to-disk backup job to a SATA volume should never impact the performance of your mission-critical databases or business applications.
Unlike SAN islands, a centralized SAN can only function with clearly defined policies and processes for requesting and categorizing storage, and when the appropriate storage tiers are assigned based on requirements.
"With storage changes impacting a large number of users, we had to adjust and tighten our storage change management policies, as well as standardize the default size of requested LUNs to 64GB," reports Carl Follstad, manager, university data management services at the University of Minnesota in Minneapolis. Putting policies and processes in place is a crucial piece of any SAN consolidation project, and it can be a challenging task, especially for smaller, less process-oriented organizations.
Finally, SAN consolidation isn't complete without a storage vendor review. If you have more than two vendors, you should work toward getting down to two vendors or even a single vendor. Having multiple vendors requires training your staff on multiple sets of equipment and tools, and overcoming the hurdle of moving data among different vendors' equipment. A single vendor strategy, on the other hand, bears the risk of vendor lock-in and losing price negotiation power, which requires more vendor management effort. "Typically, we see cost saving[s] as high as 20% to 30% in multivendor bake-offs," reports Robert Passmore, research vice president at Gartner Inc., Stamford, CT.
From a high level, all consolidation strategies can be traced to two fundamental approaches: physical consolidation by migrating to fewer, larger systems; and virtual consolidation by bringing smaller systems under a single management umbrella. In most consolidation projects, there's a blend of the two approaches.
One of the key strategies when consolidating storage is to move toward larger switches and directors, reducing the total number of switches in the SAN. Besides simplifying switch management, fewer switches reduce the number of inter-switch link (ISL) data transfers and, as a result, decrease the debilitating impact of ISL bottlenecks.
There are two options when moving toward larger switches: Deploy relatively inexpensive high-port count switches such as Brocade's SilkWorm 4900 and QLogic Corp.'s SANbox 5200 and 5600, which scale up to 64 ports; or move to director-class switches. The 64-port switches typically come in a stackable form factor and aren't designed with the redundancy features of directors; however, if deployed in pairs, they constitute a high-availability switching solution that's especially attractive for smaller and midsized SANs requiring fewer than 64 ports.
For SANs that require the highest level of redundancy, director-level switches are the ideal consolidation platform. They're available with up to 512 4Gb/sec FC ports and, with an aggregate backplane capacity of more than 1Tb/sec, are equipped to cope with the most demanding switching needs. With a passive backplane, redundant blades, switching processors, fans and power supplies, they display the highest level of redundancy, eliminating all single points of failure. But a director's high availability, performance and scalability also make it the most expensive switching solution.
Although a high-port aggregation switching solution will be part of most consolidation projects, especially to beef up the SAN core, existing switches will be redeployed rather than replaced in most cases. Legacy switches are typically moved further away from the core toward the edge, where they perform less mission-critical switching chores, connecting back to the core for inter-switch data transfers.
Repurposing or combining multiple smaller switches within a single SAN has become a viable approach as a result of enhancements in FC switching equipment. The biggest concerns with connecting isolated SANs have been problems such as rogue apps in one SAN affecting performance and availability in other SANs; FC is a basic Layer 2 protocol that doesn't have domain-isolation capabilities. Fortunately, today's FC switches, routers and directors provide features that address these concerns.
To start with, FC routing permits multiple fabrics to be connected without merging them, allowing only specific devices to communicate across SAN boundaries rather than enabling unrestricted communication. While Brocade and McData Corp. coined the term Logical SAN (LSAN) and depend on multiprotocol routers to connect separate fabrics, Cisco Systems Inc. offers its Virtual SAN (VSAN) that uses FC frame tagging to segment physical fabrics into VSANs. Another way to look at LSANs and VSANs is by comparing them to zoning: While zoning provides isolation within a single fabric, LSANs and VSANs are analogous to zones that span multiple SAN fabrics.
|Pros and cons: SAN consolidation approaches|
FC routing is the least intrusive and least disruptive way of overcoming some of the challenges involved in connecting disparate fabrics. Fabrics in a logical SAN retain fabric services like name servers, zoning databases, routing tables and domain IDs, eliminating the necessity to resolve conflicts.
"We created a VSAN on a Cisco MDS 9509 Multilayer Director and attached a departmental [EMC Corp.] Clariion FC4500 array, replacing two 16-port 1Gb Brocade switches," says Follstad at the University of Minnesota. "The nice thing about VSANs is that we were able to isolate the FC4500 array without impacting our main SAN, but still had it under single management."
Performance is a key metric and the consolidated SAN needs to ensure that all apps and servers perform adequately. There are two approaches to SAN performance management. First, a SAN can be designed to meet certain performance requirements by connecting servers and arrays to the same switches and switch blades, dedicating ports to arrays and servers where needed and limiting the use of oversubscription. The second approach uses QoS to enforce performance requirements. Features like traffic classification, limiting traffic via traffic shaping, zone-based QoS, assigning more credits to certain switch ports and bandwidth reservation can ensure a required performance level. However, the more QoS features are used, the more complex SAN management will be. To avoid unnecessary complexity, Follstad decided not to use QoS, at least at this point in his consolidation project. "Instead, we designed the storage area network to have no ISL oversubscribed more than 4:1," he says. A SAN consolidation strategy obviously needs to strike a balance between the two approaches to meet SAN performance objectives.
Storage array consolidation
Storage arrays are the core component of your SAN. They're home to your data and applications, and the purpose of a SAN is to network them into larger, more manageable and scalable storage pools. Array consolidation attempts to combine isolated arrays in SAN islands into fewer, larger physical or virtual arrays under a single management umbrella, creating a central storage pool from which LUNs get assigned to servers.
The size of the LUNs you provision is an important strategy consideration. On one hand, LUNs shouldn't be too large because bigger LUNs decrease the overall storage utilization of your SAN. But if the LUNs are too small, there may be too many LUNs, which can adversely impact array performance and make storage management more complex. To make matters worse, many arrays don't support--or array vendors may not recommend--resizing LUNs, which requires assigning additional LUNs to servers each time server volumes need to be extended. Although it largely depends on the specific environment, the "sweet spot" for the size of assigned LUNs appears to lie between 32GB and 128GB.
The easiest first step to SAN consolidation is to upgrade old arrays that are difficult to expand. Taking array utilization, projected data growth and support costs into account, upgrading may be the most economical option, especially if you have fully appreciated legacy arrays that can be combined into fewer, better performing, easier and more economically managed units. When contemplating storage array upgrades, you should also consider the aggregate savings related to the power and rack space gained by squeezing more data into smaller arrays with larger drive sizes.
Among your array consolidation options, you should consider taking advantage of larger drive sizes by replacing smaller sized drives with higher capacity drives and combining high-performance drives with slower high-capacity drives within the arrays. The larger your consolidated SAN grows, the more important it is to have more than one storage tier. In corporate reality, there will always be department heads and project managers who are eager to deploy their own little SAN islands rather than using and paying for centrally allocated storage. For a centralized storage strategy to succeed, being able to offer high performance and low-cost, large-capacity LUNs to your users is imperative. In fact, without the flexibility of multiple storage tiers, small storage islands will proliferate and a centralized corporate storage strategy is more likely to fail.
The University of Minnesota's Follstad was fully aware of this, and decided to offer his users three tiers of storage: Tier-1 storage using EMC Symmetrix DMX arrays with FC drives, Tier-2 storage on Clariion arrays with FC drives and Tier-3 storage on Clariion arrays with SATA drives. With Tier-3 storage internally priced at 25% of Tier-2 storage and one-sixteenth of Tier-1 storage, Follstad can compete with the attractive acquisition cost of isolated SAN islands. "Tier-3 LUNs comprise almost 50% of the available SAN capacity and are popular for archiving, backups and research data," he reports.
One of the biggest pain points in SAN consolidation projects, and with tiered storage in general, is data migration. The challenge of getting data from one array or tier to another, with the least degree of user disruption, increases with heterogeneous source and target arrays. To migrate data between systems of the same array family, you can use replication or the migration tools that come with the arrays. Follstad used EMC's SAN Copy to get data from a Clariion FC4700 to a Clariion CX700 array. It copies complete LUNs at FC speed. "SAN Copy is a great tool to copy massive amounts of data," says Follstad.
Storage managers have been creative when it comes to data migration. From simple host-based file copies (see "Host-based replication"), array-based replication or data migration utilities, storage managers have been struggling with the data migration challenge. "Data migration is the No. 1 reason for storage virtualization deployments today," reports Gartner's Passmore. Follstad shares a similar sentiment: "Although I don't see a need for virtualization for day-to-day storage management tasks, it would make SAN frame migration really easy," he says.
A storage consolidation analysis wouldn't be complete without looking at virtualization as a means of combining smaller storage arrays into a large virtual storage pool that's managed by a virtualization engine. The more smaller, heterogeneous storage arrays you have, the easier it is to make the case for storage virtualization. Besides the obvious data migration benefit, storage virtualization provides a single storage management platform for provisioning and managing your SAN storage. On the downside, however, virtualization isn't cheap and is still going through a maturing process with EMC, Hitachi Data Systems Corp. and IBM Corp. promoting three fundamentally different approaches to virtualization.
"The ultimate storage consolidation goal is having a single storage system that serves multiple purposes and has multiple personalities," says Tony Asaro, senior analyst at the Enterprise Strategy Group, Milford, MA. It needs to serve high-performance as well as inexpensive large-capacity volumes, block-based and file-based optimized storage, and should work with well-known protocols such as FC, IP, CIFS and NFS. Although this lofty storage goal is usually only partially realized today, virtualization seems to have the greatest potential to make this consolidated storage nirvana a reality in the not-so-distant future.