Printer Friendly

Breakthroughs In Enterprise Backup Solutions For HAS bile Servers, Part 2.

This article is the second in a two-part series. The first part appeared in the March issue of CTR.

You've probably heard a lot about the advantages of third party copy, also known as extended copy or serverless backup, in Fibre Channel SANs. Many storage pundits have touted third party copy as the "killer app" in SANs. Well, NDMP has supported its own form of third party copy since version 2.0, released over three years ago.

This solution, which is illustrated in Figure 1, is also known as filer-to-filer backup. This is a combination of the first two solutions. One of the filers has a local tape drive or automated tape library attached to it. The other filer backs up its data by reading the data off its disks and then sending it across the LAN to the first filer. Again, the NDMP client residing in the backup workstation initiates and coordinates this entire operation.

This solution trades off an increase in the congestion on the LAN for a reduced number of backup devices to acquire and manage. Another limitation is that this solution impacts the filer through which the backup data flows, even though this filer is not performing a backup. As with the first two solutions, this is most appropriate for a network that includes only a few small NDMP-compliant NAS devices.

SAN Backup Solution For Filers

In May 2000, as part of the Open Storage Networking (OSN) initiative, Quantum/ATL and Network Appliance announced a solution for backing up Network Appliance filers using a Fibre Channel SAN. This solution is illustrated in Figure 2. Just as Fibre Channel SANs, in general, evolved to consolidate backup resources by allowing them to be shared among multiple SAN-attached disk systems, the SAN Backup Solution for Filers evolved to allow multiple filers to share a single large automated tape library. In other words, this solution resolves the issues that are created when trying to scale the filer-dedicated tape backup solution to multiple filers.

This solution offers significant advantages over the previously described solutions. When the SAN backup solution was announced, it was the first and only solution that could be truly scaled to support enterprise-class storage backup requirements. However, today, with the recent advent of the NAS Backup Solution for Filers, which is described in the next section, IT managers now have another solution to seriously consider.

The promise of a SAN Backup Solution for Filers is that the backup SAN can now be shared not only by multiple filers, but by other servers and other disk systems in an open SAN. Because of the continued interoperability issues within Fibre Channel SANs, however, it is often difficult to deliver on this promise. IT managers purchase NAS filers for a number of very good reasons, including the expectation that dedicated, single-purpose NAS storage devices will be highly reliable. A SAN backup solution that is dedicated to filers is highly reliable; a SAN backup solution that backs up filers and is extended to back up a wide variety of other servers and disk systems requires significant qualification and certification efforts.

NAS Backup Solution For Filers

In January, 2001, Quantum/ATL announced the product availability of the second major OSN initiative solution--the backup of filers over Gigabit Ethernet. This solution, which is also known as the NAS Backup Solution for Filers, is illustrated in Figure 3.

Architecturally, this solution takes advantage of NDMP's third party copy capability. In that sense, it is conceptually similar to the NDMP third party copy solution with a couple of notable differences.

First, the NDMP server functionality is embedded directly into the automated tape library. What this means is that instead of relying on a NAS filer to "front" for the library and convert the NDMP file data to SCSI for the library, the library can do all this for itself. From an NDMP perspective, embedding this intelligence into the tape library raises the backup device--the "sink" for the data--to a peer level with the NAS device--the "source" of the data. This gives the library complete independence to back up any of the NAS filers without impacting the other NAS filers on the network. The benefit of this independence is that all of the backup devices, including the individual tape drives, can be dynamically shared among the filers. The other benefit is that it is now possible to easily scale this solution simply by adding more backup devices to the network.

Second, all the backup data can be routed over a separate dedicated Gigabit Ethernet backup network. This avoids congesting the client LAN. In practice, the IT manager creates this second network by integrating another Network Interface Card (NIC) into each filer. The library and all the filers are then added to the backup network. Note that this step is similar to what is done when creating a Fibre Channel SAN, except that one adds NICS to the filers instead of Fibre Channel host bus adapters (HBAs).

The similarities between the NAS and SAN backup solution don't end here. If you compare Figures 2 and 3, you can see that this solution is really a "SAN backup solution" which transfers NDMP file data over GbE instead of block data over Fibre Channel. Because of this, the NAS Backup Solution for Filers provides all the benefits of a Fibre Channel SAN solution. These benefits include device sharing across the network, LAN-free backup, serverless backup, extended distances, and simplified cabling, no practical limitation on the number of devices, and zoning ("virtual LANs") for security. As described in the next section, even the data transfer bandwidths of the NAS and SAN backup solutions are very comparable. Additionally, the IT manager doesn't need to add a second fabric--a Fibre Channel fabric--to his existing Ethernet fabric. It is possible to realize all these benefits plus leverage the existing Ethernet network hardware, software, knowledge, training, and experience. This is truly a significant breakthr ough in enterprise backup.

Data Transfer Bandwidth

The NAS Backup Solution for Filers transfers data from the filers to the automated tape library using file-oriented NDMP data over TCP/IP. The SAN backup solution transfers block data from the filers to the library using SCSI over Fibre-Channel. Due to the lower protocol processing overhead of Fibre Channel, it enjoys much lower latencies than TCP/IP for round trip request/response pairs. This is necessary for block level data transfers to achieve high throughputs. NDMP, on the other hand, was written for data streams as opposed to block level requests, and from its inception has been designed for efficient use over networks. NDMP implements a "sliding window" algorithm on the data stream to effectively mask the latencies inherent in TCP/IP, analogous to the way modern CPUs use pipelines to mask memory latencies.

How do these factors balance out in the real world? In a series of tests, we found that, up through an overall solution throughput of 50MB/sec, the NAS solution performed at least as well as the SAN solution. We determined these results while backing up a Network Appliance 840 filer running Data ONTAP release 6.0.1 with a Quantum/ATL P3000 library populated with multiple DLT7000 tape drives. We conducted these tests with both compressible and non-compressible data.

Although the data throughput is comparable for the two solutions, the filer CPU utilization is higher in the NAS backup solution, and the CPU becomes saturated when the filer throughput exceeds 55MB/sec. In other words, filer CPU saturation does not limit throughput unless the filer is trying to transfer 2:1 compressible data concurrently to more than six DLT7000 tape drives. If the data is less compressible than this, then the number of tape drives to which each filer can stream data at full tape drive bandwidth increases.

Note that the throughput at which the library can "sink" data from the filer scales linearly with the number of tape drives in the system. This is because the NDMP server functionality is distributed throughout the tape drives in the library. Therefore, by adding more drives to the library, the library will never be the bottleneck on the overall system throughput.

Above 50MB/sec, the Fibre Channel SAN solution demonstrates better throughput because of lower CPU utilization. However, we anticipate that with the adoption of jumbo frames there will be significant increases in the NAS solution efficiency effectively removing the performance differences above 50MB/sec. The jumbo frames will have the effect of reducing the TCP stack overhead, and therefore, the CPU utilization, by a factor of six because the frame size increases from 1,532 bytes to approximately 10,000 bytes.

The capabilities of NAS file servers are continuing to evolve at a phenomenal rate, and they are being more widely deployed throughout the entire range from workgroups through enterprise data centers. Fortunately for IT managers, the evolution of backup solutions for these appliances has kept pace to offer data protection capabilities that are commensurate with the needs of the end users over this full range. Figure 4 illustrates how each succeeding backup solution continues to offer increased capabilities and benefits to the end user.

The most dramatic breakthroughs in backup solutions for NAS filers--both the SAN and NAS backup solutions--have been developed in the past year and were announced as part of the Open Storage Networking initiative. Both of these backup solutions are targeted at the enterprise customer. They offer high reliability, significant overall solution performance, and reduced management costs through consolidation and sharing of the backup devices.

In the NAS backup solution, the data is transmitted over TCP/IP, which in turn, runs over Ethernet. This offers significant total cost of ownership benefits to the IT manager. Ethernet based libraries are a perfect product match for NAS filers; users with large investments in filers by definition have already made large investments in their Ethernet infrastructure. While the rest of the storage world is still defining the details necessary to deploy complete and robust systems for transferring storage data over Ethernet, NAS file server owners have already started to enjoy the full benefits of using Ethernet both for accessing primary data and for backup.

Furthermore, as Quantum/ATL demonstrated at the Comdex show in November 2000, when these other standards are fully defined and more widely accepted, the Ethernet library that supports NDMP over TCP/IP today will be able to be easily upgraded in the future to accommodate iSCSI or other Ethernet-based protocols. Steve Morihiro is the CTO and vice president of engineering and Rory Bolt is the director of systems design at Quantum/ATL (Irvine, CA).
Figure 4: Evolution of Filer Backup
 NDMP Data High Device Low
 Required? Protection Performance Sharing TCO $
 Backup YES
 Over LAN
Tape Backup
 Party Copy
COPYRIGHT 2001 West World Productions, Inc.
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2001, Gale Group. All rights reserved. Gale Group is a Thomson Corporation Company.

Article Details
Printer friendly Cite/link Email Feedback
Title Annotation:Technology Information; NAS file servers
Author:Bolt, Rory
Publication:Computer Technology Review
Geographic Code:1USA
Date:Apr 1, 2001
Previous Article:Gadzoox Introduces 2Gb Switch.
Next Article:Without Management There Is No Infrastructure.

Related Articles
OSN Demonstrates Data Backup Over GbE! But is it NAS or SAN?
Mercom Intros Support Option.
NAS/Tape Backup Strategies Cope With New Data Cornucopia.
Breakthroughs In Enterprise Backup Solutions For NAS File Servers.
Design Considerations In Enterprise Storage Networks.
SANs VS. NAS: What You Should Know And Why You Should Care.
Speeding up the network: D2D backup lets VARs beat the Bottleneck Bugaboo. (Nectivity).
The new world of NAS virtualization: the next generation could fulfill NAS' promise. (Storage Networking).
SAN and NAS convergence: can satisfy storage requirements while reducing complexity and overhead. (Storage Networking).
Hardware beats out software for top honors.

Terms of use | Privacy policy | Copyright © 2021 Farlex, Inc. | Feedback | For webmasters