Storage Consolidation

On a nuts-and-bolts level, you can use centralized storage and fabric switching to allow "any to any" connectivity. Multiple platforms, servers and environments can access resources in a shared storage pool on SANs or high-end NAS systems. Ideally, you can design and implement your setup to share logical data stores across platforms. For example, legacy AIX and MVS applications might query and update the same inventory database. And administrators can allocate additional storage resources from the shared physical environment as needs change.

Most important, though, is centralized storage's zero downtime. The problems that plague distributed data storage are unlikely to occur in a centralized environment. Fewer overall storage devices, more robust management systems and centralized backup can translate into 99.999 percent uptime.

With DASD (direct-access storage device) arrays, redundant fabric switches, robust backup devices and storage-management software, your distributed-platform site can attain mainframe-level storage integrity and reliability. These technologies have been delivering in the big iron world for years--and you'll find storage vendors are more than happy to offer the goods for your Windows or Unix platforms.

Of course, all this reliability and performance comes at a higher per-megabyte cost than direct-attached drives (thanks to all the gee-whiz components), but the outlay can be offset if your situation is right. All vendors will be able to provide reliability specs for their products; be sure to get customer references in your industry so they can prove it to you.

Zero downtime is probably the most compelling reason to centralize storage, but there are less obvious cost benefits. Decreased storage reserves can generate substantial savings. All shops keep an available stockpile on each of their storage environments. A site might have 500 servers, with each one connected directly to a 50-GB RAID. The usual modus operandi is to maintain an average 20 percent reserve in each environment to handle short-term growth. It's a smart strategy, but it means a lot of unused storage on each box.Moving from distributed storage to a single environment greatly reduces the need for a surplus. Instead of a separate pool on each box, you'll have a smaller, centralized store for any of the connected environments. Voil, just-in-time storage allocation. Administrators can set standard threshold alarms for capacity limits and allocate storage from the centralized pool as needed.

Automation tools reduce the significant labor costs associated with storage-provisioning tasks, according to a Yankee Group report. What's more, centralized storage is easier to manage than distributed storage once it's up and running, thanks primarily to fewer moving parts. With fewer devices and more robust management tools and flexibility, your IT staff can concentrate on more pressing issues.

McKinsey & Co. says SAN and NAS solutions yield economies of scale that can reduce staffing costs by up to 80 percent. Shops with large-scale SANs manage storage four to eight times more efficiently than those using conventional methods. So if your site has four people managing the storage farm, it may take only one person to deliver the same level of support post-centralization.

At a minimum, a happily humming SAN or NAS implementation will free staff from the drudgery of storage management. No more never-ending tape-swapping, RAID-drive replacement or dry eyes from reading capacity monitors. By consolidating your data-center storage, you move all your RAID stacks and tape loaders into one area. When it's time to swap a drive or load your DLT, everything is right at hand, rather than in a locked cabinet halfway across your production floor.

There are plenty of real-life success stories. Recently, a capacity-management type told us his storage centralization tale. His company, a Fortune 500 financial services firm, was looking at a large server refresh (more than 1,400 boxes) in its primary data center.The company's prior operating model had left it with old, out-of support DAT or DLT autoloaders on every creaky Wintel server in the shop. The firm's final design gave it a roughly three-to-one server consolidation, with all data storage moved to a Fibre Channel SAN with centralized backup. Instead of buying 400 autoloaders for the new boxes, the company built up its existing SAN (which until then served only the mainframes and a small subset of midrange environments) and improved its reliability and performance. The savings realized from not buying hundreds of new autoloaders paid for the lion's share of the company's new DASD. Other savings resulted from reduced physical-plant requirements.

Moving the SAN out to all the systems not only freed up floor and rack space, but also reduced power and HVAC demands. As a bonus, the smaller footprint meant the crowded data center didn't need to be expanded.

Any company running a large 24/7 transactional system on distributed systems should centralize storage using a SAN. The savings from avoiding downtime or loss of data easily makes this move cost-effective. For sites with deep pockets, the added bandwidth of Fibre Channel over iSCSI may be worthwhile, yielding faster queries and more transactions per minute. In a smaller environment or at a site with 9-to-5 application needs, a NAS system usually makes more sense (see "ROI/TCO Examples,").It may sound too good to be true, but it's exactly what your management wants to hear. Vendors, industry studies and those in the trenches have all come back with the same advice: A well-implemented SAN or NAS requires fewer person-hours than conventional direct-attached storage. Depending on your inclination, this can reduce head count, improve service and increase storage per administrator.

Rather than laying off people, you can temper staffing plans to reflect the efficiencies of centralized storage. The same number of folks you have managing 2 TB in your direct-connect world may be able to manage 4, 8 or 13 TB down the road as you grow in a centralized environment. Break down your staffers' duties to the task level to determine where they'll save time and you'll have the metrics you need to project future savings.

Let's start with some basic concepts and apply them to a conventional storage environment. Drives are cheap--and since price per unit will only get cheaper, we can plot a flattening curve for hardware costs associated with your planned growth in storage. Salaries, however, are expensive. We tend to want high-quality people managing our environments, so we compensate accordingly. Unlike hard drives, good employees aren't getting cheaper.Recognizing that one person can manage only a fixed quantity of servers and direct-attached storage, calculate the threshold points at which you'll need additional staff to support growth. Your staffing-cost curve becomes a step model, with a big jump each time someone is hired; as you move into the future, staffing costs will likely increase faster than hardware costs.

What could be driving those staffing costs? Assuming some form of direct-attached backup device (DLTs, for instance) each server will need, at a minimum, care and feeding in the form of tape swaps. Since each server has its own internal and/or external RAID, the number of individual devices under management (controllers, drives) increases with each additional server. With each additional component, the overall complexity of your shop increases, and management becomes more challenging.

Direct-attached storage makes sense in a small shop: If you're plotting 100 percent growth over five years, your dozen servers and 2 TB might still be manageable when they blossom into 24 servers and 4 TB. But if you're running 400 servers with the same projected growth, how many more folks will you need to stay on top of 800 servers and their routine maintenance, break-fix and upgrades? Growth is good for the company, but it can be rough on the operations staff.

That's where the benefits of centralized storage come in. All "care and feeding tasks" are greatly reduced, thanks in large part to a reduction in the number of moving pieces. Instead of monitoring RAID controllers in each server, you need only sit on top of your SAN console. Data purge and archive, disk maintenance and other routine activities can be managed as a whole rather than server by server. Centralized backups can save hundreds of staff hours per month (ignoring the additional potential benefits of online backup and increased application uptime).

Centralized storage can also enable more nimble responses to storage requests, improving user (and management) perceptions of your performance. In a direct-attached environment, each server and application requires solid capacity planning for future growth. Unfortunately, despite the best strategy, unexpected storage demands can have your operations staff scrambling to add capacity at a moment's notice. Unplanned activity or an ill-tempered application can choke a standalone storage environment, leaving you with a passel of unhappy users.One way to mitigate these risks is to set triggers and alarms. You also can minimize them by influencing user behaviors and maintaining sufficient buffers of excess capacity on each server. A well-designed NAS or SAN will let operations staff dynamically allocate additional storage in response to any of these issues. Instead of scrambling to stay one step ahead of

the storage ceiling on each of your DAS boxes, a centralized solution will let your staff work more fluidly with business and applications teams to manage spikes and surges.

So look hard at your existing metrics. If you don't have good numbers on hand, start keeping track. If a large chunk of your staff's time is spent walking the aisles of your farm to swap tapes, add drives or replace toasted components, take notice. If you're not using a solid monitoring/management package to stay on top of capacity and performance stats, start tracking how much time is spent baby-sitting each box. If you know big growth is coming, start working your numbers and talking up that you'll be able to "do more with less." CFOs just love that phrase.NAS

Compared with direct-attached storage, network-attached storage gives you a leg up in terms of management tools, integration, cost-effectiveness and reliability. NAS systems are packaged as easy-to-install "appliances." Think of them as dedicated servers with operating systems specialized for storage.

Connecting to your LAN and using remote console management, these solutions are typically seen as a standard file share on the network. NAS enables a file system to be handled remotely (usually via CIFS or HTTP), with redirector software sending a file request over the network to the centralized storage device to serve up data. More robust solutions scale as requirements grow via modular construction, and they can provide NFS (Network File System) and SMB (Server Message Block) connections. Depending on your performance requirements and budget, you can outfit your NAS with dual NICs, Gigabit Ethernet and ATA/IDE, SCSI or Fibre Channel drives.You can improve reliability by going with multiple power supplies and redundant cooling fans. Backup solutions can be direct-connected or network-based, depending on your needs, usage patterns and LAN bandwidth.

Best of all, you can have a basic NAS up and running in minutes. Ease of management is a strong selling point, as is the low price--solid solutions can be had for $5,000 to $30,000. For small and midsize sites, NAS can be a simple, easily justified centralized solution that doesn't require you to venture into the SAN world. (For more details, see "First-Class NAS".)

SAN

A storage area network is dedicated solely to data storage and retrieval. Its main components are storage devices and an independent network. The storage components include fast disk drives and large-scale tape backup devices, with storage-management software tying it all together. Network hardware may include switches (Fibre Channel in FC SANs, Ethernet in iSCSI) that let components talk to one another, as well as connectivity bits such as HBAs (host bus adapters), which connect external devices to the SAN.

Ideally, the switching fabric makes it possible for storage devices from multiple vendors to interconnect and function with multiple operating environments--letting, say, IBM AIX, MVS, Sun Solaris and Windows 2000-hosted applications access common physical storage environments.SANs rely on either Fibre Channel technology, which provides high bandwidth (2 Gbps) linking storage components, switches and HBAs, or iSCSI technology based on standard IP topologies. SANs were initially proprietary Fibre Channel architectures; but iSCSI is the new kid on the block, generally offering a lower-cost, if somewhat slower (1 Gbps) alternative. Either option can incorporate legacy SCSI devices. There may be some performance degradation with legacy drives, but the ability to maximize existing capital investments can strengthen the business case.

A solid design should rely on dual (redundant) switches with ISL (interswitch links) and dual HBAs to provide maximum uptime for connections between storage subsystems. (For more information on the technical pros and cons, see "iSCSI SANs"; and "High on Fibre".)

There's a lot of talk about proprietary versus open standards for SAN architectures. In an ideal world, a Fibre Channel SAN would work with storage and network components from any vendor, letting you integrate legacy storage and new components as needed. The FC SAN industry is not quite there yet; CIM/OM (overlapping with Bluefin) and SMI-S are the leading FC standards. The Storage Networking Industry Association--whose members include Brocade Communications Systems, Cisco Systems, EMS, Hewlett-Packard, Hitachi, IBM, McData Corp. and Nortel Networks--continues to work on storage interoperability. Meanwhile, the IEEE is weighing in with its Storage Systems Standards Committee. The iSCSI vendors, for their part, are also pursuing interoperability--the Internet Engineering Task Force has a working committee dedicated to IP Storage specifications, producing RFC 3347 back in July 2002.

If you need the 2-Gbps speed of Fibre Channel, expect to limit your FC implementation to a subset of vendors whose equipment has been proven to play well together. When talking through RFPs (requests for proposals), be sure to get references from customers detailing their environment and legacy components.

Bottom line: SANs are expensive and not easy to implement. Few shops have any in-house FC experience, and few training centers or schools offer FC curricula. Thanks to basic economics, you can bet that any FC consulting you bring in-house will be dear. Because iSCSI relies on conventional IP, this type of SAN will be more familiar to your staff. But you're still likely to need outside help.When discussing price per megabyte with any vendor, get a detailed breakout of the cost model and make sure the model matches your situation. Remember, dollar-for-dollar benefits may not be there for your site. Be sure to shop around. While a good 16-port entry-level FC switch will set you back $20,000, a realistic base capital expenditure, including switches, HBAs and cabling, can easily cost $50,000--and that's just to set up the subsystems in preparation for storage.

Robust FC SAN implementations can cost hundreds of thousands of dollars, and they continue to go north as requirements increase.

Going with iSCSI for your SAN will save money over FC; FC carries a 50 percent premium per port over copper Gigabit Ethernet ports, exclusive of optics. ISCSI allows you to spend less, thanks to less expensive components and more familiar technology. But you'll still be making a substantial investment building out a new storage infrastructure. You don't want to run your iSCSI SAN on the same TCP/IP LAN that carries your normal network traffic (see "Don't Sink Your IP SAN"). You'll also need to invest in TOEs (TCP off-load engines) and/or HBAs for any substantial iSCSI installation.

Be sure to review feature sets, and remember what your goals are in moving to any SAN--solid performance, redundancy from bulletproof components, clean hardware design, functional management software and flexible backup. Determine what best suits your needs and shop accordingly.

NAS provides an excellent, relatively inexpensive solution for small shops. FC SANs deliver performance for those who can justify the high cost of implementation. ISCSI SANs fall somewhere in between, leaning more toward the FC model than the NAS one in terms of cost and performance. If you need the reliability of a SAN, want to utilize your existing IP know-how and are satisfied with the Ethernet switching speeds, iSCSI may be the best solution for you--and may likely yield the best business case after you run the numbers.The benefits of centralized storage management really come out as the size and demands of the enterprise grow. The bigger your shop, the more demanding your customers and clients, and the more convoluted your existing storage schemes. Therefore, the more you can benefit from centralized storage.

The technology is cool, but money matters more. Emphasize revenue gains, increased service levels, happy customers and reduced expenses. After all, these are the benefits of storage centralization, regardless of which method your company chooses.

Joe Hernick is an IT director with a Fortune 100 company. He has 12 years of consulting and project-management experience in data and telecom environments. Write to him at [email protected].Let's compare the costs and benefits of traditional distributed storage with those of a SAN. We'll do this for two companies. One is a small shop with a 15/6 production cycle; the other is a midsize, 23/7 enterprise.

To keep things simple, we're assuming comparable growth rates, hardware and staffing costs, and reliability stats for both. The main expense difference is the "cost" of system downtime: The bigger shop incorporates a larger hourly figure for outages, reflecting a larger hypothetical client base.

All numbers are conservative. For instance, only 10 percent of the "soft" benefits have entered into the ROI calculations. (The numbers would quickly tip in favor of SANs if you weighed these benefits more heavily.)The following examples are only a guide. You'll need to build your own cost-projection model, taking into account expenses and savings specific to your organization, including those associated with outsourcing or training.

Most vendor sites offer ROI tools if you don't feel like getting intimate with Excel. The "Savings/Expense Considerations" listed below can help you as well.