The technology is evolving slowly, but the benefits take you one step beyond LAN-free backup.
By Heidi Biggar
After months of being billed as the first "killer app" for storage area networks (SANs), server-less backup/recovery implementations are beginning to take form, albeit at a much slower pace than many had expected. Key ingredients, such as third-party copy and special control software, are making their way into various hardware products and backup software applications. Nonetheless, analysts don't expect robust server-less capabilities before year's end-and those could be cost-prohibitive to all but enterprise-class applications.
"Server-less backup is still a glimmer in people's eyes," says Paul Mason, vice president of infrastructure hardware at International Data Corp. Yet, the promise of freeing up application servers from performing time-consuming backup functions holds their attention.
"User's don't care how it's done. They just need to get their backup done while their systems are running other applications," says Jeff Vogel, vice president of marketing at McData.
In a recent survey conducted by International Data Group, 84% of respondents said they thought SANs could improve performance, while 82% believed SANs would aid in disaster recovery (see "Users warming up to SAN benefits," InfoStor, January 2000, p.1). Server-less backup should assist in both these areas.
Server-less backup/recovery has all the benefits of LAN-free SAN implementations, with a noted addition: it tackles the issue of server I/0. Because backup data flows directly from disk to tape, the application server's role is reduced to that of custodian of the backup application and control software (e.g., Legato Celestra Power Agent). The copy agent or data mover (e.g., SCSI Extended Copy), which sits on a bridge, router, or other device, moves backup data across the SAN. Servers are then free to handle other applications with virtually no backup interference.
By eliminating the server from the backup process, users realize immediate performance and -potentially-cost benefits. Throughput is now determined by the speed of the storage devices, not the processing power of the server. As such, the backup window virtually disappears and productivity and service levels improve significantly.
As for reducing overhead, the potential certainly exists; however, the gains may not be quite as immediate as initially expected. The idea is that by using inexpensive copy devices, instead of high-end servers, to move data, overhead is reduced. Additionally, since control software (e.g., Celestra) is loaded onto existing application servers, users are spared the costs of purchasing and maintaining dedicated tape servers. Of course, this assumes that software prices-at least initially-won't negate these savings.
The Celestra Power Agent issues copy commands to a bridge/router or switch (with embedded scsi Extended Copy support) to move block lists of data directly from disk to tape.
On the downside, several hurdles must be overcome before these benefits can be realized. Foremost, there's the pressing issue of hardware and especially software availability. "Right now, we've got an "n" of 1, where "n" equals Legato, when it comes to software choice, and the rest of the pieces [SCSI Extended Copy standard, NDMP support, etc.] are still coming together," says Tom Petrocelli, Atto's Fibre Channel product manager.
"No vendor wants to be perceived as the in-hibitor of new technologies," adds Sean Derrington, an analyst with the Meta Group consulting firm, "but server-less is a software play." As such, primary responsibility lies largely with the software vendors (e.g., Computer Associates, Legato, Tivoli and Veritas) to bring products to market first. Nonetheless, some hardware vendors such as Chaparral, Crossroads, and SmartSAN have re-cently introduced products that support server-less backup, specifically Celestra.
Product availability concerns aside, other issues worth noting include:
- Scalability. A server-less implementation is not scalable if the source (e.g., disk array) and destination (e.g., tape library) devices are attached to a single router. Further, a router's performance capabilities inherently limit scalability. Put simply, a router has enough bandwidth to service a finite number of drives.
For example, "Mammoth-2 drives, with an aggregate throughput of 12MBps, would likely max out a single router (i.e., the drive-to-router ratio is 4:1)," says Tim Weir, senior product manager, Exabyte's storage automation and solutions division.
- Support. Closely related to the issue of scalability is support. Initial server-less implementations will likely lack broad platform support and some vendors may only support proprietary backup applications.
- Manageability. Unlike traditional back-up approaches, server-less backup re-quires an end-to-end view of all con- nected devices for proper management. This level of management is not yet available.
- Recoverability.Recoverability is argu-ably the most controversial issue surrounding server-less backup/recovery. The premise is that some vendors only address the backup side of the equation, not backup and recovery. Though such claims appear to be unfounded, in some instances recovery processes are not as efficient as backup procedures.
How it works
There are a variety of ways to achieve server-less backup. However, the most common approach combines Legato Celestra 2.2 (to be released in the second quarter), an NDMP-compliant backup application (e.g., Legato Networker or Veritas Netbackup), and a third-party copy agent (in this case, SCSI Extended Copy) to move data directly from disk to tape with virtually no server involvement-something other techniques (e.g., data replication or clustering) can't claim.
"I hesitate to call the other approaches 'server-less backup,'" cautions Scott McIntyre, business line manger for storage networking at Legato. "They have the merit of mov ing data off the application server, but they transfer that load to a different server." Celestra, specifically release 2.2, takes a different approach. Although Celestra Power Agent sits on the server that needs to be backed up, it does not consume server bandwidth.
Celestra, not third-party copy, is the "brains" of the application. (Celestra is based on code Legato inherited from Intelliguard last year.) "Our model was to make the copy command as simple and lightweight as possible, while putting all the intelligence in the power agent," says Don Trimmer, senior technical strategist at Legato and chair of the SNIA backup working group.
The process is actually quite simple. The Celestra Power Agent takes instructions from the backup application (via NDMP). It then determines what needs to be backed up and sends that block list to the third-party copy agent (SCSI Extended Copy), which is embedded in some type of intelligent device (e.g., a bridge, router, hub, library, disk array, or controller).
The data mover then reads the data from the disk array and writes it to the tape device. After the copy device has finished moving the data, the power agent reports back to the backup server (see figure).
Legato expects to release Celestra 2.2 next quarter, carrying a fairly hefty price tag: $22,425 for the Celestra Power Agent, $15,000 for each terabyte of data to be backed up, and $22,425 for NetWorker.
Celestra 1.1.2 and NetWorker 5.6, which include NDMP support, are available. Release 1.1.2 enables server-less backup in a workstation environment via Legato's Celestra Copy Agent, not SCSI Extended Copy. Supported tape drives include DLT7000, 9840, and 3590.
Though Legato has been most vocal about its server-less plans, other vendors are also in development. Computer Associates, for example, will demonstrate its capabilities at CA World next month, and Veritas expects to announce product in the fourth quarter.
"Being able to leverage the Image Option in ArcServe gives us an advantage," claims Peter Malcolm, senior VP, business management for storage, Computer Associates. "Image Option not only enables users to back up data at very high speeds at the block level, while bypassing the operating system, but it allows you to restore data at the file level," Malcolm contends.
Server-less backup is part of CA's SANiti initiative. The functionality is expected to be integrated into the ArcServe Enterprise Library Option (ELO). The first release will not be NDMP-compliant.
Veritas is taking a slightly different approach. Though the company is reportedly developing its own "power agent," it is also working on other technologies that will enable users to back up their data more efficiently.
"We're going to offer this support in two products: our own third-party copy agent in Netbackup 4.0 [to be released in the fourth quarter] and another form of Frozen Image Backup, which uses an alternate server as the third-party copy agent," says John Maxwell, director of product management, backup solutions, Veritas.
By building APIs into the Netbackup 4.0 file system, Veritas claims its server-less capability will provide a level of recovery assurance unmatched by other vendors. "You need API-level integration with your file system and operating system to guarantee this works properly," says Maxwell.
Meanwhile, Tivoli says it will enable server-less backup via SANergy (formerly Mercury SANergy), TSM 3.7, and TDP for workgroups.
On the hardware front, users will soon have a variety of options. At this early stage, the most common approach among vendors is to embed SCSI Extended Copy into bridges, routers, or other SAN devices. Most of the devices are being tested with Celestra, several of which-including products from Chaparral, Pathlight, and SmartSAN-have reportedly passed the Celestra CCTT test.
For example, Crossroads recently launched a new line of Fibre Channel-to-SCSI routers. Two of the three models-the 4250 and 4450-support Extended Copy. The routers feature LVD Ultra2 SCSI connectivity and FC_Tape support to aid in error recovery. ADIC plans to integrate the routers in its Scalar 100 and 1000 tape libraries. Atto also has plans to offer Extended Copy as a software option to its FibreBrige Enterprise.
Chaparral has announced the DM1000, a SAN appliance (or data mover) that enables tape-to-disk, disk-to-tape, and disk-to-disk data movement, adding to its existing line of storage routers. The DM1000 includes embedded Extended Copy support. The unit plugs into a switch or hub to provide server-less data movement anywhere in the SAN, not just between disk and tape.
What's the next step? "You need to look at the performance and scalability of these products," says David Padilla, product manager for HP's software & SAN management operation. He says that if you're trying to design a large SAN configuration with multiple storage devices and multiple operating systems, a switch or router might not be the answer, but some other type of "black box."
Third-party copy is a generic term used to describe the mechanism for moving data on behalf of another application. SCSI Extended Copy Command, on the other hand, is a specific example of a third-party copy technique.
Based on code developed by Crossroads, Intelliguard, and Pathlight, the Extended Copy command is expected to become an industry standard for data movement. Its development was spearheaded by the SNIA backup working group, and the specification is currently in the ANSI approval process.
With roots in network-attached storage (NAS), the Network Data Management Protocol (NDMP) is a high-level interface that drives the Celestra Power Agent. NDMP does not lock users into specific backup applications.
"NDMP is a standard way to hook things together," explains Don Trimmer, senior technical strategist at Legato and chair of the SNIA backup working group. "Users can make buy decisions for frameworks and clients independently."
Server-less backup benefits enterprise applications
By Seth Asser
Some companies are turning to bandwidth-boosting technologies such as ATM and Gigabit Ethernet to help solve the enterprise backup/restore requirements of multi-terabtye very large databases (VLDBs) and server farms. However, while these enhanced network topologies do make more bandwidth available, they are limited by the processing power of individual servers, making them a short-term-and often costly-fix.
More bandwidth doesn't solve the issues of effectively managing, backing up, and restoring huge volumes of data, especially when many distributed disks, each attached to a different server, are involved. What's really needed are service-level agreements and around-the-clock data access that doesn't affect production servers. One answer: LAN-free, server-less backup, the benefits of which are many:
- Server-less backup is specifically designed to support I/O-intensive backup operations while operating in parallel to the LAN, without interfering with the bandwidth required for regular data communications.
- Because server-less backup centralizes the backup function and storage resources, numerous backup applications-which affect the CPU performance of the application servers-are not needed. This cuts product costs, not to mention training overhead.
- A single backup and recovery interface streamlines administration and management. The consolidated backup approach also allows IT managers to implement cohesive backup strategies that reflect enterprise-wide service-level agreements, management policies, and disaster recovery options.
- Server-less' single interface facilitates data recovery, enabling database administrators to quickly find and restore data to its original location. The server-less implementation should feature an automated discovery process that handles logical-to-physical mapping at the storage subsystem level, without user intervention, and prior to each backup.
- Some server-less implementations feature bulk data movement between disk and tape for optimum backup and recovery performance. Also, data replication capabilities can eliminate the need to back up production volumes. In many cases, service-level agreements can only be met by making addressable copies of data on storage subsystems. This is particularly important as single instances of databases increasingly exceed multiple terabytes.
- In some implementations, server-less backup enables administrators to consolidate backup and recovery operations regardless of operating environment. In such cases, it provides a platform for centralizing the storage requirements of server farms, thereby eliminating the need for locally-attached storage and minimizing the costs and headaches associated with administering and managing numerous independent disks.
- In addition to resolving backup and recovery issues associated with proliferating VLDBs, server-less backup can effectively address the storage management problems associated with expanding Windows NT and NetWare networks.
As the applications that run server farms become more critical, the need for a more reliable, automated, online, and centralized approach to backup and recovery-one that cost-effectively addresses enterprise-wide concerns and policies for data protection-becomes increasingly important.
At the same time, the backup solution for a distributed storage environment must address the same business continuance issues associated with backing up VLDBs, which include policy-based management, centralized storage, a single interface, point-in-time replication of data, remote mirroring, and cluster support.
Seth Asser is senior product manager at EMC (www.emc.com).