Standard eases lab's backup problems

Like many federal organizations, the Los Alamos National Laboratory in New Mexico found that network-attached storage (NAS) disk arrays were a handy solution for adding new storage capacity without the expense or hassle of adding new servers and server-attached disks.

Unfortunately, they also discovered that their original setup for backing up NAS-based data to tape was not so ideal. Because all of the backup data was funneled through a dedicated server running the backup software and connected to a tape library, it created a network bottleneck and provided little flexibility for devising better setups when computing needs changed, as they often do at Los Alamos.

The answer to the lab's problem lay in products that support an increasingly popular standard called Network Data Management Protocol. In a nutshell, NDMP takes the server running the backup application out of the data stream, making the process much more efficient and flexible.

Beyond the added efficiency, there are other benefits. NDMP enables vendors to offer buyers a range of interoperable products from which to choose. The protocol also supports numerous backup configurations, giving information technology shops like the one at Los Alamos more deployment options and protection for their storage investments.

"We change our architecture quite often, so a top benefit for us [of NDMP-based backup] was that we wouldn't have to change these large pools of [NAS-based] data storage every time the architecture changed, because they could support different modes of backup," said Curtis Canada, a systems scientist with the Advanced Computing Laboratory at Los Alamos.

Such benefits are among the reasons many storage vendors are adding NDMP support to their products. Among this growing list are backup software packages such as BakBone Software Inc.'s NetVault, which is used by Los Alamos; Veritas Software's NetBackup; Computer Associates International Inc.'s BrightStor and many others (see box).

Also in on the act are NAS device manufacturers, including Auspex Systems Inc., EMC Corp. and Network Appliance Inc., as well as tape library vendors such as Quantum Corp., Sony Data Systems and Spectra Logic Corp.

If there is any shortcoming to the current NDMP market, it's that support for the protocol is stronger in the high end of the market filled with data center-class products, according to Steve Kenniston, an analyst with the market research firm Enterprise Storage Group.

"Support is not as good when you drop down to the next tier, where you're talking about a huge installed base of [Microsoft Corp.] Windows devices," he said. "Some of the big backup vendors have not developed products for this market yet because they are waiting to see whether Microsoft will bundle" NDMP functionality into its products.

Solving NAS Backup

The NDMP standard is hardly new. Network Appliance and PDC Software (now part of Legato Systems Inc.) began co-developing it 1996. From a basic functional perspective, the idea was to create a standard mechanism to retrieve data from one storage system, copy it directly to another one and restore lost data if needed.

Having a standard way to handle these processes saves storage and backup vendors the effort of designing their products to work on different platforms. For example, Network Appliance uses a proprietary operating system for its popular line of NAS boxes. NDMP helps backup software vendors who want to appeal to as many customers as possible by supporting current versions of the Network Appliance operating system.

Indeed, Los Alamos' purchase of BakBone's NDMP-compliant backup software was spurred by its desire for a better way to back up its eight Network Appliance F800 NAS boxes, also called filers. The filers, which are connected to users' workstations and one another by a Gigabit Ethernet network, store huge datasets created by the lab's experiments in global ocean modeling and the like.

Backing up that growing pile of data by sending it from the eight NAS boxes through a single server connected to a tape library would have become more difficult in the limited time available for backups, Canada said. (Another common way people back up multiple NAS boxes is by connecting a tape drive to one of the NAS boxes, but again, bottleneck issues can arise.)

The NDMP-compliant backup products provide Los Alamos with an alternative. Now, the lab has two Spectra Logic tape libraries, each with 16 drives, connected directly to the Gigabit Ethernet network, not to a dedicated server as before. A Linux server is still needed to run the BakBone backup software, but its role is not as the main data conduit as it was with the server before.

"It's more of a traffic cop, directing the backup job and describing where the data is stored on tape," Canada said. "But it's not installed in the data flow, so we're not drawing cycles from it during backup."

There are other advantages. For one, the lab has created a virtual storage pool out of the two tape libraries, enabling a feature called dynamic drive sharing, Canada said. "That makes things very easy for us, because all of the tape drives are accessible by any" of the NAS boxes, he said. "We don't have a complicated array of what [device] can talk to what."

Giving users this kind of flexibility is what NDMP is all about.