This article can also be found in the Premium Editorial Download "Storage magazine: Overview of top tape backup options for midrange systems and networking segments."
Download it now to read this article plus other related content.
Here's a classic storage manager's nightmare: Suddenly end users can't access 100GB of critical data because of a bent pin on a SCSI connector. And here's what really makes you wake up in a cold sweat in the middle of the night: There's no backup copy.
That happened at SAS Institute a few years ago and it ignited an effort to revamp how backups were done. Backup at the Cary, NC, software company is serious business - at stake are millions of lines of code and dozens of simultaneous software builds.
Kelly Wyatt, a senior systems programmer at SAS, knows these challenges all too well. Her team, which supports more than 4,000 people involved in maintaining the 47 individual applications within the company's product portfolio, is continually looking for better ways to manage the company's nearly 8TB of live production data.
The immensity of SAS' storage needs made it clear to Wyatt and her team that simply relying on backups by trying to coordinate independent local tape systems was never going to provide adequate manageability. The firm installed filers from Sunnyvale, CA-based Network Appliance, which not only made highly scalable file servers, but also had worked closely with Intelliguard - now part of Legato - to develop the flexible backup management protocol, Network Data Management Protocol (NDMP).
Although initially designed to improve backup of NetApp filers, NDMP has taken on a wider role: Most network-attached storage (NAS) vendors have endorsed
"It's about ensuring that you're backing up everything that needs to be backed up," says Jay Desai, data protection solutions manager with NetApp. "This is a universal backup agent for all applications. If you have a lot of IP investment, this is a complement to that solution just like Fibre Channel [FC] attached tape drives are in a SAN. That's one of the reasons people implement NDMP - to leverage their investment in large tape resources."
NDMP eliminates the need for data to be funneled from remote locations through a central backup server, then dumped into a single backup repository. Initially, it enabled centralized management of multiple local tape drives in each filer. Newer versions allow you to use centralized tape repositories. Built in a client/server model, NDMP has hooks in both backup management software - the client - and virtual state machines built into compliant networked storage devices - the servers.
At SAS, IT uses Veritas' NetBackup to manage backup of 18 NetApp filers, which provide 11.5TB of NAS space used to store data from Windows and Unix systems spread across multiple logical volumes. Some 30 locally attached DLT 7000 drives are involved, with four drives on each of two NetApp filers dedicated to backing up data from non-NetApp file systems.
Wyatt explains the impracticality of the alternative - going to each filer, typing DUMP and then trying to make sense of all the tapes. "We offsite hundreds of tapes a week, and don't want to have to hand label what they are. With NDMP, we are able to very effectively and efficiently back up 8TB of data on 18 filers, and it's really helped us with consolidating data on fewer servers."
NDMP's expanding role
As NDMP has evolved, its expanding scope has forced a redefinition of terminology. A NDMP client is now also referred to as a Data Management Application (DMA), while a NDMP server is now also called a Data Service Provider (DSP). Various DSPs have been defined, including data (e.g., servers with internal storage, NAS devices, SAN-attached storage subsystems), tape (jukeboxes, general servers with shared tape drives or writable CD-R libraries) and SCSI pass-through DSPs. The upcoming NDMP v.5 adds a Translate DSP, including N to M stream multiplexers and data compression.
Building a storage strategy around NDMP can deliver faster performance rather than relying on the indirect paths taken in conventional backup models. This is because NDMP streams - which only carry a small volume of management-related information - direct the data flow between servers and their locally attached storage devices. If you opt for locally attached tape, data itself isn't transmitted over the corporate network, eliminating the congestion that can degrade application performance and push backups perilously close to the edge of their backup windows.
If you're interested in centralizing tape in a large library, NDMP v.2 and v.3 support sophisticated three-way backups. These allow NDMP-compliant filers to direct their data to backup devices that may be connected to a different filer altogether. By combining this configuration with a restore from attached storage, a user can automatically schedule redundant backups of stored data.
For example, data from one filer-attached tape drive can be shuttled to another tape drive using NDMP. Three-way backups allow data to be pushed to tape jukeboxes, from one filer to another - allowing maintenance of online backups instead of pushing data to tape - or from a filer without local tape storage to one that's connected to a library. Data can be sent from multiple filers to a single tape library, or split across multiple, individually attached tape libraries to increase backup performance.
Because it's built around TCP/IP and not the expensive FC connections required for distributed SAN backups, NDMP is an easier and faster approach that delivers similar functionality into the NAS environment. That makes it a godsend for companies that have struggled with the personnel burden of expanding backup tasks as their corporate data grows.
"We have quite a few techniques for heterogeneous backup and recovery on the SAN side," says Mike Adams, product marketing manager for NetBackup with Veritas. "The question is, do you want to have a box that's very easy to manage? People choose a NAS device because they want file sharing that's simple and easy; SAN folks decide they want to do block-level sharing of information. Depending on your business needs, that can be more technology than you want to deploy."
Support for Windows and Linux
Because NDMP is now supported in Windows CIFS (common Internet file systems) and Unix NFS (Network File System), it can back up data from both environments while retaining each one's file attributes and other idiosyncrasies. This capability led Honeywell Satellite Systems (HSS), Glendale, AZ, to implement NDMP along with an upgrade of its tape drives intended to speed backup of its 700GB Auspex NS2000 file server.
This was first published in July 2002