WAN optimization and disaster recovery

WAN optimization offers a perfect fit with disaster recovery by squeezing more bandwidth from links between primary and secondary data centres.

WAN optimization is a natural choice for disaster recovery. That's because one of the most common methods of putting a disaster recovery provision in place is to build a second mirror site at a location remote enough that it's unlikely to suffer the same disaster striking the primary site.

In such a setup, recovery readiness consists of backing up all or part of the enterprise environment across sizeable distances -- and this makes huge demands on wide-area links. As a result, WAN optimization products have become a popular method of squeezing more bandwidth from expensive low-bandwidth connections while also lowering latency.

The ability of WAN optimization to dramatically reduce data volumes when replicating across a wide-area link is often the main justification for deploying it, according to Clive Longbottom, service director of business process analysis at research firm Quocirca. "It's hard to justify WAN optimization solely on the basis of a disaster recovery project, so users often justify it with business continuity benefits. But this means that disaster recovery also gets the benefit."

What is WAN optimization?

To have a mirrored second site ready for use in case of disaster means copying data from one data centre to a second facility. The problem is that replication takes time -- and the further away the two facilities are, the longer it takes.

Throwing bandwidth at the problem doesn't necessarily provide a solution. Latency -- which is a consequence of the finite speed of light -- leads to saw-toothing, so called because that's what the bandwidth graph looks like when data rates move up and down as they peak, max out the connection's bandwidth, then fall back as systems at either end try to resend data. "The key is to smooth out data rates and remove peaks and troughs so that throughput and link utilisation are maximised," Longbottom said.

WAN optimization technology achieves this by lowering the volume of data and its latency using the following four main techniques:

  • Data deduplication: Identical in principle to the data deduplication with which most storage managers are familiar. It stores an index of all data transferred and ensures that any piece of data is transferred only once. The result can be bandwidth reductions of up to 90%.
  • Protocol optimization: Reduces the protocol overhead associated with TCP by manipulating the packet size. The gains are smaller but significant.
  • Application acceleration: This manages the communications of particular applications, such as MAPI, CIFS and SQL, reducing their impact on the WAN via a variety of techniques, including caching.
  • Compression: Some vendors apply compression, which can squeeze slightly more data down the pipe.

Disaster recovery provision can take advantage of all of the above techniques. Another feature of WAN optimization products is the ability to optimize mixed traffic, which is useful as disaster recovery traffic often has to compete with production data as well as backups and replication.

WAN optimization vendors

Four vendors dominate in the competition to deliver WAN optimization technology suited to disaster recovery: Cisco, Expand Networks, Riverbed Technology and Silver Peak Systems. Of these, research firm Gartner puts Riverbed first in terms of technology with Cisco close behind.

Cisco, however, is not placing much focus on developing its WAN optimization products, according to Longbottom. "That there seems to be the arrogance to think that Carrier Ethernet will come along fast enough and at a price point low enough to make all organisations of all sizes avoid the need for WAN acceleration is somewhat frightening," he said. "However, Cisco is bought by big enterprises because it's the safe bet."

Cisco offers three Wide Area Application Engine (WAE) appliances aimed at data centre deployments: the WAE-674, the WAE-7341 and the WAE-7371. They differ mainly in the volumes of data they can store and manage, with the WAE-7371 including up to six 300 GB hard drives and 24 GB of RAM.

Riverbed's range of data centre-oriented Steelhead appliances -- the 5050, 6050 and the 7050 -- offer capacity from 2 TB to 5 TB, and up to 48 GB of RAM.

Silver Peak Systems is not a significant player in the UK, but it focuses on the high end and disaster recovery applications, with the smallest of its seven NX Series appliances, the NX-1000, capable of optimising links of up to 4 Mbps, and the largest, the NX-9000 with 8 TB of storage, managing up to 1 Gbps.

Expand Networks' Accelerator appliances focus on application acceleration, offering three product series with bandwidth management that is application and protocol specific. The three ranges cover the data centre, branch offices and mobile workers, and have a maximum bandwidth of 250 Mbps.

Also in the WAN optimization market is NetEx, whose HyperIP products claim a maximum bandwidth of 622 Mbps.

All of these vendors offer WAN optimization technology that boosts bandwidth and cuts latency. While Riverbed Technology's hardware may be more capable, discussions on industry forums suggest that the key difference between it and Cisco boils down to complexity. Cisco products are consistently accused of being overly complex as well as difficult to deploy and manage, while Riverbed's products are more automated and easier to drop into the network. Cisco supporters argue that Cisco technology is a better fit in an environment where Cisco products predominate and that it costs less than Riverbed's offerings.

WAN optimization boosts global DR plan

One UK-based, global law firm (which asked not to be named) found that WAN optimization technology allowed it to run disaster recovery processes across a 45 Mbps link between its UK and Hong Kong data centres, despite experiencing annual data growth of approximately 50%.

The company wanted to replicate its storage arrays between the two sites but, prior to the deployment of WAN optimization, high latency meant the link carried more retry requests than real data.

Upgrading the link would not have solved the latency problem and would have been very expensive. "We needed WAN optimization; there was no alternative," a spokesperson said.

The problem disappeared when the firm added a pair of Silver Peak NX-7600 appliances at either end of the link. They were then able to successfully configure a pair of remotely connected NAS boxes as LAN-attached, the devices experiencing no communication issues despite being several thousand miles apart.

"Now the bottleneck is the LAN, not the WAN," the spokesperson said. A big influence on product choice was the NX-7600's ability to accelerate all traffic, not just specific applications. Without this, it "would have caused all sorts of problems trying to figure out where any problems are," the spokesperson added.

The problem of moving ever-growing volumes of data around the organisation is set to stay with us as the trend toward consolidation into fewer, larger facilities continues. Meanwhile, long-distance links remain expensive and suffer from high latency. WAN optimization allows WAN access to feel like LAN access, while also improving disaster recovery performance.

However, in the longer term, the general trend toward Carrier Ethernet and a growth in the number of WAN providers that include optimization as part of the service bundle could see the decline of WAN optimization as a discrete product category.

Read more on Disaster recovery