The key to controlling SAN implementation costs is to reuse existing hardware and software.
By Tom Petrocelli
As more and more data is required to be continuously available, the efficiencies of shared- storage architectures are becoming necessary for some companies. This is especially true for Internet businesses, where millions of customers expect to have order history and inventory on-line, all the time. It is even more important for industries such as energy distribution and financial markets, which have legal requirements for high levels of data availability. Waiting hours for information stored off-line on removable media to be made available to customers or internal users is no longer acceptable. Storage area network (SAN) architectures can help manage the large amounts of continuously available data.
Unfortunately, new technology is not usually easy to implement. Problem areas include personnel training, technical hurdles, and time constraints. In the case of storage networks, specific issues need attention. Financial considerations-especially the initial cost of implementing a new network infrastructure-coupled with the hesitation to replace functional equipment that has not been fully depreciated, are often of primary concern. Of even greater concern is the perception that a massive data migration project is required.
Don't Replace, Reuse
A simple solution to these problems is upgrading, rather than replacing, major system components. By reusing components, financial and time goals can be met while reducing overall risk.
Frequently when new technology is introduced, the choice is to deploy or not to deploy. This is not the case with Fibre Channel SANs. Many elements of the existing storage infrastructure can be upgraded or "front-ended" with proxy devices to take advantage of SAN architectures. This is especially true of the most expensive and difficult-to-maintain devices.
By reusing major components, a greater return on investment (ROI) is realized on existing devices due to an extended life span. Reusing existing devices also means greater leverage of existing IT training and experience. This reduces SAN implementation costs and provides an overall cap on total cost of ownership (TCO).
Figure 1: In this upgrade example, the original configuration is based on a 100BaseT LAN backbone.
Finally, by keeping existing storage devices intact, the need to migrate data from one device to another is reduced or eliminated. This greatly reduces the risks of implementing a new architecture because data is left in its original location.
While all SANs are different, there are some common elements. This is where opportunities for upgrades exist:
- Servers, workstations, and other computers.
- Storage devices, which are often the most expensive element. Large disk arrays, as well as tape libraries, are typically deployed in SANs. Smaller arrays can also be used for workgroup applications.
- Storage network infrastructure, including hubs, bridges, routers, and switches. These devices offer the least opportunity for upgrade and reuse because they would not have been deployed in previously implemented storage architectures.
- System software, especially operating systems. Until recently, most operating systems and applications did not support the shared-storage model of SANs. However, systems software is a critical element because the potential for data corruption or overwriting data is high if SANs are deployed without some form of data management.
- Application software, especially applications that access data directly. Many applications were designed in an environment where they had total control of the data. Any shared-data applications accessed data through a medium such as a file server, which regulated access to the data on the disks with policy-oriented controls.
Many of the elements of the legacy storage architecture can simply be upgraded or con nected through proxy devices to a Fibre Channel SAN. Some areas of the SAN architecture offer better upgrade opportunities than others.
Server hardware is the easiest and least costly major system component to upgrade. Simply replace the SCSI host bus adapter (HBA) in the server with a Fibre Channel HBA. While the server can now access more devices, they will still look like SCSI disks and tape drives. This is possible because most Fibre Channel adapters provide drivers that transport SCSI-FCP (SCSI Fibre Channel Profile).
There are, however, limits imposed by most operating systems on the number of devices and logical units that can be accessed through a SCSI driver. Some of these limitations may be overcome with new drivers.
The most troublesome part of implementing a SAN involves the storage devices. However, by using proxy devices, existing storage can be attached to the Fibre Channel SAN at a relatively low cost. A proxy device allows one type of subsystem, such as a parallel SCSI or IDE disk array, to look like a Fibre Channel device. These proxy devices are called bridges, or routers, because they provide a bridge between the two physical architectures while using the same upper-level protocol (SCSI).
Figure 2: Combining a LAN and a SAN addresses data availability, connectivity, and network congestion issues.
Typically, the proxy device costs much less than the storage device. Many manufacturers of large-scale storage devices have embedded bridges to provide SAN-ready capability.
Operating system support for SANs is woefully inadequate. Until the launch of Windows 2000, the Intel-based server market did not have SAN support. Even support within Windows 2000 is limited. Some Unix platforms have better capabilities, but most are still not up to current requirements. The most important facility that many operating systems do not provide is the direct access to shared disk space without overwriting or corrupting the data on the disks.
However, third-party software can enhance operating systems by providing policy-based file and volume management software. While the emerging category of SAN operating system software has promising features, a simpler file or volume management approach is sufficient for many applications.
When low-cost LANs were first introduced, many application software packages were unable to operate in the networked environment, and the same is true for SANs. While many applications have been SAN-enabled, many are still not ready for this environment.
Most backup software now has a SAN option, often available as an upgrade to existing packages. Database software is also reasonably SAN-aware. And major tape backup applications are now almost entirely available in SAN-enabled versions.
Some vertical application areas that would benefit from SANs, such as video editing, are only now seeing software that is SAN friendly. Unfortunately, these often require replacement of software and are not simple upgrades. Still, there has been considerable success in this market by coupling existing video software and hardware with volume or file-management software.
As an example, consider a site with customer relationship management (CRM) applications such as sales, order entry, and customer service (see Figure 1). There is a need to tie this system to an Internet on-line store to provide sales and customer service.
This configuration is based on a 100BaseT Ethernet backbone. Connected to the backbone network are Unix servers running databases, an order entry system, CRM and call tracking system, and order processing. Also on the LAN is a telephony service that routes call information to the CRM system, and a Windows NT-based tape server running backup software. Attached to each server is a 300GB disk storage subsystem. Additional Unix servers are used as firewalls and Web servers.
The IT department has identified three problem areas centered on data storage systems:
- The Internet store. The transaction load is heavy, and is bogging down the LAN as customers attempt to get service from the order entry and customer care systems.
- The backup system. With large increases in the volume of stored data, the nightly backup window has increased to a point where it interferes with normal operations.
- Integrating the data. The company wants to tie all of its data together. In addition, Web customers have come to expect all order and service-related data to be available whenever they want it. The IT department does not always have access to data, and the amount of data crossover between different databases is saturating the LAN.
These problems can generally be characterized as data availability (as opposed to system availability) and connectivity issues. Fibre Channel SAN technology can help solve these problems.
In our example, IT believes that by having the same data stores available directly to the servers via Fibre Channel, rather than through the LAN, their applications will be easier to deploy and scale. In addition, by having data available to more than one server, the chances of a server failure crippling the system and making data unavailable to customers is greatly reduced. To accomplish these goals, IT recasts the current architecture into one that combines a SAN and a LAN (see Figure 2).
Cost, however, is a chief concern because the IT department does not have an unlimited budget. As such, they choose to conserve much of their budget by upgrading system components wherever possible. The major upgradeable components were the tape library, backup software, database software, servers, and disk arrays.
The IT department adds Fibre Channel switches, hubs, and cables, which are necessary to build the storage network (see Figure 3). These components, especially the switches, represent the greatest expense.
Figure 3: A fully configured SAN might include Fibre Channel switches, hubs, and host bus adapters as well as Fibre Channel-to-SCSI bridges/routers.
To upgrade the servers, SCSI HBAs are replaced with Fibre Channel HBAs. The tape library and disk arrays can also be easily upgraded by using Fibre Channel-to-SCSI bridges. With bridges in place, the tape library and disk arrays appear to the network as Fibre Channel devices, not parallel SCSI devices.
There is another benefit to the upgrade: additional intelligence. A bridge may have additional software (such as failover and security options) that the original tape library or disk array may not have.
Finally, the backup software and databases are easily upgraded with SAN-ready versions.
By reusing most of the existing infrastructure, the SAN can be deployed much faster. In addition, training and experience already invested in major components are leveraged for overall lower cost. Subsequently, while some software may need to be updated, major homegrown applications do not need to be rewritten. This speeds development and helps control costs.
This strategy enables IT departments to provide better service, with less disruption and cost, in a fraction of the time it would take to develop and build new systems.
Tom Petrocelli was formerly a product manager with Atto Technology Inc. (www.attotech.com), in Amherst, NY.