OSN Demonstrates Data Backup Over GbE! But is it NAS or SAN?

This article is one in an ongoing series related to the newly formed Open Storage Networking (OSN) initiative. Please see the June issue of CTR ("OSNI Puts FC's Feet To The Fire," page 1) for a complete discussion of OSNI

How can resource-strapped IT departments contend with shrinking backup windows, skyrocketing data, and the cost of new storage technology? Simple. Just combine the performance and connectivity of SAN with the ease of deployment, manageability, and cost benefits of NAS and Gigabit Ethernet (GbE). That was the solution proposed at the recent Networld+Interop show by Network Appliance and Quantum/ATL, who jointly announced the Open Storage Network (OSN) initiative, which proposes truly open, standards-based storage networking for continuous access and availability of data within network environments. While NetApp, Am, and their partners already offer a true Fibre Channel SAN-based secondary storage solution for NAS file servers, it was the OSN GbE solution that caused decidedly more stir at the show.

The OSN GbE secondary storage solution involves backing up data from NAS file servers such as the industry-leading NetApp filer, across GbE to an automated tape library. Today, GbE is utilized in high-speed Local Area Networks to transport data in file format between servers and clients or between file servers and application servers. More recently, GbE is receiving considerable interest as a possible alternative to Fibre Channel as a SAN transport medium, due primarily to cost considerations. As an example, a draft specification for SCSI over TCP/IP is posted on the Internet Engineering Task Force (IETF) Web site. Additionally, other vendors have proposed various protocols for transmitting storage data in block format across Ethernet. In this article, we will examine the benefits of the OSN GbE solution and attempt to answer the question--NAS or SAN?

The Ease Of Use Of A HAS Device

The primary technology underlying the OSN GbE solution, illustrated in Fig 1 and Fig 2, is the open Network Data Management Protocol (NDMP) standard (www.ndmp.org). NDMP allows a network backup application to back up and restore the data on a NAS file server without being installed on the file server itself. This is a necessity because NAS file servers typically do not run standard operating systems. Without NDMP, network backup applications must be ported to all the different NAS operating systems. NDMP solves this problem by separating the control and the data transfer functions of the backup/restore operations, as shown in Fig 3. The control components such as scheduling, backup database, etc., reside on a standard Solaris or Windows NT workstation. This is called the "NDMP client." All of the components required to support movement of data reside on the "NDMP server(s)."

The low-bandwidth control communications between the NDMP client and the NDMP server(s) occur across a TCP/IP connection. Similarly, high-bandwidth data communications among two or more NDMP servers occur across a TCP/IP connection, which can be separate from the network used for control communications.

To date, the NDMP server functionality has been embedded solely within the NAS file servers. This is how data is currently backed up from the file server to a direct SCSI-attached tape drive or tape library. This is also the means by which data is backed up from multiple file servers to the OSN Fibre Channel SAN-attached tape library.

The most significant difference between the two OSN secondary storage solutions is that with the OSN GbE solution, the NDMP server functionality is distributed by being embedded directly within the tape library. From the NDMP client's perspective, the tape library and tape drives are all available to be shared across the network for backing up the NAS file servers. In the same way that embedding the file system within the NAS file servers provides enhanced independence between the storage resources and the application servers, embedding the NDMP server functionality within the tape library provides enhanced independence between the backup resources and the file servers. In both cases, the benefits are the same--enhanced sharing and access to data resources and enhanced scalability.

The backup data is transferred between the NAS file servers and the tape library in file format across the TCP/IP connection. This is an enterprise-level NAS backup solution because the transfer is done in file format, rather than block format, from a very narrow technical perspective. More importantly, to the end user, this solution offers all the benefits that make NAS file servers so popular, including ease of use and the ubiquity of TCP/IP and Ethernet. Additionally, this solution utilizes existing network hardware, software, and people-ware infrastructure, creating a very attractive cost consideration.

Many Of The Benefits Of A SAN Solution

The OSN GbE solution is clearly not Fibre Channel, nor does it transfer data from the file server to the library in block format. So it is definitely not a SAN solution from a technology perspective. However, let's look at the backup solution from the end user's perspective and see how well it addresses the benefits that are commonly associated with the SAN backup model.

* Device sharing across the network. One of the key benefits associated with SAN is the ability to share backup resources among multiple SAN-attached disk systems. This enables end users to consolidate backup resources. The OSN GbE backup device is readily shared among multiple NAS file servers because the library has the NDMP server functionality embedded within it. In fact, because of the enhanced independence of storage resources from the server operating systems and file systems, the OSN GbE solution offers true dynamic sharing of backup resources. In a SAN, each of the operating systems and file systems in a heterogeneous server environment will try to claim all the primary and secondary storage resources. This is typically solved through the partitioning of storage devices or by establishing zones throughout the SAN.

* LAN-free backup. As shown in Fig 2, the OSN GbE solution provides a separate backup network through which high-bandwidth backup data flows from the file servers to the library. This backup network is distinct and independent of the client network used to transmit data between file servers and application servers. Since the backup data does not flow through the client LAN, the backup operation does not impact its bandwidth, which is the objective of LAN-free backup.

* Server-less backup. Server-less backup, also known as third-party copy or extended copy, is a feature in which the backup operation is accomplished without affecting the I/O bandwidth of the application server. It is interesting to note that the Fibre Channel SAN server-less backup concept evolved from the NDMP concept and you can see they share many similarities. In the OSN GbE solution, the application server is insulated from all of the backup data and, therefore, every backup is by definition server-less backup. The NAS file server, on the other hand, performs all of the high-bandwidth transfer of backup data to the library and, therefore, accessibility to its data is limited while the backup operation is being performed. However, this same limited accessibility to data exists when a SAN-attached disk system is backed up through server-less backup.

* Extended distances and simplified cabling. Compared to parallel SCSI systems, Fibre Channel and the OSN GbE share the advantages of extended distances and simplified cabling. The actual distances that are possible are a function of the specific media (copper versus optical, etc.). Serial communications are used in Fibre Channel and Ethernet, thereby greatly simplifying cabling.

* Limitation on number of devices. Again, compared to parallel SCSI systems, neither Fibre Channel nor OSN GbE impose a practical limitation on the number of devices that can be networked together in a storage network.

* Data transfer bandwidth. Although, by specification, Fibre Channel and GbE transfer data at roughly the same Gigabit per second, there is more overhead and latency associated with transferring data through GbE. This is potentially significant for backup applications. Fortunately, the NDMP protocol utilizes a "sliding window," which allows large amounts of data to be streamed without waiting for individual responses. This helps mask the effects of the latencies inherent in the underlying TCP/IP on the overall transfer rates. Early measurements have recorded over 92.8% transmission efficiency, using the NDMP protocol over TCP/IP. Additionally, this is a situation that continues to improve as costs for processor and communication bandwidths continue to decline.

* Zoning. Zoning is required in SAN systems to overcome the issue of various operating systems and file systems trying to claim all the storage resources in the SAN. Zoning is not required, even if it is possible by implementing "virtual LANs," because this problem doesn't exist in NAS systems. Zoning can also be used to limit access to storage resources for security or policy reasons. In the OSN GbE solution, zoning can be implemented using virtual LANs for security purposes. Additionally, the NDMP protocol allows the user to use passwords to restrict access to the network storage devices.

What's The Answer--NAS Or SAN?

Steve Duplessie, senior analyst and founder of The Enterprise Storage Group, recently made this plea to a searchStorage.com chat audience: "Let's stop calling it NAS or SAN. Let's just call it networked storage."

Of course, what he meant is that, while concrete distinctions exist--at least for now--between NAS and SAN, we need to focus on the combined benefits of these technologies to the end user. The OSN GbE solution takes a "best of breed" approach to secondary storage. It provides the ease of deployment and use of NAS file servers and eliminates the problems associated with Fibre Channel by utilizing GbE as a transport medium. The OSN GbE solution is a natural fit for those who currently use NAS systems and Ethernet infrastructure, since it allows users to further leverage their investments in Ethernet hardware, software, and personnel. Plus, it provides the heightened performance, availability, and accessibility to data promised by true SAN systems.

As the industry continues to look creatively at other ways to solve the end-user's data storage challenges, other solutions will surface that continue to blur the distinctions between NAS and SAN. The real question is: what benefits will the solution offer to the end user?

Steve Morihiro is the chief technical officer and vice president of engineering and Rory Bolt is the director of systems design at Quantum/ATL