Data center consolidation initiatives continue to push greater volumes of traffic onto wide area networks (WANs), and the impact of storage traffic transported between data centers is no exception. Cheaper high-bandwidth WAN links have made it easier to handle the sheer volume and high-availability requirements of WAN replication and disaster recovery (DR) traffic, but WAN managers know that bigger pipes don't solve every problem.
Data center WAN optimization vendor Silver Peak Systems and startup Infineta Systems recently announced multigigabit WAN optimization appliances, which they claim are designed to keep up with the ever-widening pipes connecting data centers for WAN replication and DR.
"One thing we can be certain of is [that] rarely does bandwidth demand drop organically," said Jim Frey, managing research director at Enterprise Management Associates. "It's really recent that this level of capacity [in WAN optimization appliances] has been a requirement because WAN links have never been this high."
Historically, demand for private high-capacity WAN links mostly came from financial institutions—the only organizations with enough compliance-related backup demands to require them and flush with enough cash to afford them—according to Joe Skorupa, research vice president at Gartner Inc.
You know the phrase, 'You can never be too rich or too thin'? Well, you can never have too much bandwidth or too little latency.
Research Vice President, Gartner Inc.
As bandwidth has gotten cheaper and compliance requirements have fanned out to other industries, more enterprises are deploying these bigger links between data centers, he said.
At the same time, data center consolidation initiatives have led to enterprises pushing bigger loads of data back and forth between data centers that are farther and farther apart from each other. That's a problem for WAN replication and DR traffic, particularly with EMC's asynchronous SRDF software, Skorupa said.
Seventy percent of enterprises upgrade the WAN links at their data centers every 18 months, according to a recent survey commissioned by Infineta of 1,000 enterprises worth at least $1 billion. Of those enterprises following this pace of growth, 69% identified WAN replication as the key driver, according to the survey.
"Storage protocols like EMC's SRDF/A start to break down over longer distances and latencies," Skorupa said. "You know the phrase, 'You can never be too rich or too thin'? Well, you can never have too much bandwidth or too little latency, so it's an ever-escalating question of how we can get more through the same pipe and how we can compensate for [the limitations of] the speed of light."
Finicky storage protocols, legacy dedupe affects WAN replication
Silver Peak has long marketed itself as the data center-focused WAN optimization vendor, but its lower-end models have found a home in branch office deployments. Not so with its latest model, the NX-10000, which is intended for WAN replication use cases, according to Damon Ennis, vice president of product management at Silver Peak.
The NX-10000 is Silver Peak's highest-end model, supporting 2.5 Gbps WAN links; 256,000 simultaneous sessions; 1.6 TB of solid-state drive (SSD) storage for deduplication; and four 10 Gbps Ethernet local area network (LAN) interfaces for 10-gig data centers. The list price is $299,995.
Storage traffic is "basically doubling every year," and compliance requirements are dictating that enterprises replicate this over greater distances, Ennis said.
"It used to be that financial organizations in New York would back up their data to New Jersey ... but now they can't do that because a terrorist attack that's going to take out New York will probably take out New Jersey as well. So now they've got to back it up to Colorado," he said.
Coupled with the increasing volumes of WAN replication and DR traffic, those demands are driving the need for higher-capacity WAN optimization appliances, Ennis said.
"The standard replication stacks from the [storage] vendors simply won't work [across those distances], so they need WAN optimization to make it feasible," he said. "And every time they increase the amount of storage they can [replicate], we need to increase the [capacity required] to replicate it."
Infineta has made similar claims about what it can do for WAN replication and DR traffic with its Data Mobility Switch (DMS)—its first product since the company launched in 2008. Infineta's custom hardware can support 2 Gbps, 5 Gbps or 10 Gbps ingress wire speeds on the same appliance; capacity increases become available through software upgrades. List prices range from $80,000 to $340,000, depending on capacity and feature set.
The DMS looks at smaller segments of data during deduplication than those looked at by other vendors' products—8 bytes versus 4 kilobytes, according to Haseeb Budhani, vice president of products at Infineta. Although it's a more efficient method of deduplication, vendors often shy away from it so as not to risk performance drops, he said.
But the DMS can process smaller chunks of data at higher levels of capacity by running its deduplication on hardware, whereas other vendors run it on software in an attempt to keep costs down, Budhani said. As a result, the average port-to-port latency is 50 microseconds, he said.
"We basically had to invent a new deduplication [technology]," Budhani said. "We're not even in the class of WAN optimization. We're in the class of a router or a switch."
Meanwhile, Silver Peak and Infineta aren't the only vendors in this space, said Gartner's Skorupa. Riverbed Technology's Steelhead appliances max out at 1 Gbps, yet they are not uncommon in WAN replication deployments, he said. F5 Networks' BIG-IP—a multipurpose application delivery controller, which includes a WAN optimization module—tops out at 42 Gbps and is another formidable contender for inter-data center deployments, he said.
Optimizing WAN replication across 1 or 10 Gbps links
Longtime managed hosting and services provider NaviSite Inc. historically had few requirements for inter-data center connectivity. But that changed upon launching managed cloud services within three of its 10 global data centers about two years ago, according to Denis Martin, CTO and executive vice president of the Andover, Mass. -based provider.
"We weren't moving gigabytes or anything near that. [Our focus] was accelerating the end-user experience—sure, with the spurious large file transfer at some point," Martin said. "But with the cloud nodes, [moving large volumes of data] is inherent to the fabric itself. We were giving customers the ability to say, 'Replicate this environment, which may be 5 terabytes, to San Jose or London."
Martin had deployed WAN optimization appliances from Riverbed and Silver Peak to address the managed services side of the house, likening the delivery to headquarters-to-branch deployments. But he sought something more "purpose-built" for the 1 Gbps MPLS connections between his cloud data centers, each of which transports a terabyte of data every four to five business days.
Martin, who deployed DMS in his data centers three months ago, said Infineta's hardware-based deduplication delivers on its promises.
"What we saw were tremendous data reductions," he said. "One of our basic tests was on a 2.15-gig file. On the naked MPLS, it took 8.3 minutes, and then it took a minute-and-a-half with DMS ... [and] with smaller files, things that took 11 seconds now take three seconds. We got the same [results] pretty linearly. It doesn't get better as it gets bigger, but it's still damn good."
Jürgen Hansjosten, CEO of Teragate AG, a cloud storage provider based in Munich, Germany, has increased the WAN links between his data centers from 100 Mbps to 1 Gbps over the past two years. He expects to push them to 10 Gbps within the next year, at which point he expects to trade in his 1 Gbps NX-9000 WAN optimization appliances from Silver Peak for the new NX-10000s.
"The speed between data centers is increasing continuously," Hansjosten said. "[The NX-9000s] are meeting my needs today, but it's to your benefit if you have multiple gigabits [of capacity available]."
Let us know what you think about the story; email: Jessica Scarpati, News Writer.
Dig deeper on Network disaster recovery