It's the final frontier in data center networks: Consolidating storage replication onto the production wide area network.
Today many large enterprises still maintain a second wide-area network or private line for storage area network (SAN) replication across data centers. The replication is essential for disaster recovery initiatives and distributed storage systems and it has traditionally required a dedicated network that is highly reliable and low-latency and delivers packets reliably and in the correct order – characteristics that are essential for the successful replication of storage from one data center to another.
Storage replication is also essential for supporting cloud computing and software-as-a-Service (SaaS) initiatives. Enterprises want to be able to deliver stored data to end users across the globe, whether the applications users are running are in a local, corporate data center or within Amazon's EC2 cloud infrastructure.
"Some people are starting to scale their data centers down," said Michael Cucchi, CTO of Par4 Technology Group, a virtualization consulting firm. "They design data centers that function at 60% to 70% of peak load and they want to be able to spin up cloud resources on demand for a period of peak processing."
As enterprises virtualize their data center infrastructure and leverage cloud resources, they need full wide area network compatibility with their storage networks and the ability to span large geographic connections, Cucchi said. "Things like desktop virtualization is made much easier and much more seamless when the wide-area network is coherent across all these parts."
Thus, enterprises are looking at the separate wide-area network or private line that they maintain for SAN replication and wondering why they don't consolidate in order to reduce costs and complexity and also enable a more coherent network strategy for cloud computing.
"Traditionally the storage network has always been kept separate from the company data network," said Bob Laliberte, storage management analyst with Enterprise Strategy Group. "And the storage guys built their own private optical networks to connect data centers in large part due to the high performance applications that they were supporting could only traverse certain distances and they need that dedicated bandwidth to be available to them."
Enterprises are now looking to consolidate their SAN replication onto the primary corporate WAN, according to Apurva Dave, senior director of product marketing for Riverbed Technology.
"This is a potential way to cut costs," he said. "Enterprises are looking everywhere and saying, 'Why do we have this duplicate network? Isn't there a better way to do this?'" Dave said enterprises are also interested in the long distances that primary IP-based WANs can transport data. This approach allows a company to replicate its data from New York City to Omaha as opposed to from New York City to New Jersey.
Dave said one Riverbed customer, Australian law firm Allens, Arthur and Robinson (AAR), recently needed to find a way to replicate 2.5 terabytes of data across EMC Symmetrix arrays.
"They worked with the engineers of [EMC], who said, 'This seems fine. You should look at investing in a 40Mbps WAN link," Dave said. "They are a law firm that is spread out across Asia, and Asia has very high latency and is very expensive. So they were going to have to spend and additional… $100,000 per year. Instead they utilized our Steelhead appliances and they found that they could usually get a 4x or 6x reduction in data. They found they were able to do the replication they desired over their existing 12Mbps link."
"When you talk about data replication or SAN extension, in many instances the WAN bandwidth can be 50% of total cost of ownership for a disaster recovery initiative," Jeff Aaron, vice president of marketing for Silver Peak Systems. "If you can leverage some cost savings on the other side of the house for storage environments, you can obviously significantly lower the overall cost for storage."
Maintaining WAN performance
Transport of SAN replication across the primary IP-based WAN, such as an Internet VPN or an MPLS network, offers plenty of challenges.
"The biggest challenge is that as you move toward a shared network, like MPLS or Ethernet VPN, while you get more bandwidth and capacity for cheaper, the quality of the WAN link decreases. That decrease quality comes in the form of packets being dropped or delivered out of order. When you're doing SAN extension or SAN replication, [the data center storage systems] require a very high sustained throughput," Aaron said. "So if you start to drop packets or deliver them out of order, that means the host starts to come out of synch and they miss their replication cycles entirely."
This consolidation of SAN replication on the primary WAN network has prompted leading storage vendors like EMC and Brocade to work closely with WAN optimization vendors such as Riverbed and Silver Peak, certifying these vendors to work with their storage products.
"Things like WAN optimization help to overcome network integrity issues to help clean up those WAN links so those cost-effective shared links actually perform like private lines," Aaron said. "So now, for the first time, you can converge your SAN extension on the same WAN link that you use for everything else."
Cucchi said that this consolidation is definitely happening, but network engineers are still isolating storage traffic across the production WAN.
"Now that can be done on a VLAN or plumbing basis or it can be done elegantly higher up in the stack, but I would say most enterprises are still isolating storage networks from production application networks through configuration and routing protocols and VLANs and tagging and things like that."
Aaron said this trend will require WAN engineers to cooperate with storage administrators more closely than ever before.
"If you are looking at doing SAN extension across the WAN then the storage and networking guys need to talk," he said. "Anything that touches the network ultimately has to get the network guys involved. We may be sold as part of a SAN extension or data replication or data consolidation project and it may be driven by the storage guys, but at the end of the day we are a networking product. You need networking guys involved because this requires IP addresses, routing policies, and QoS policies that all need to be configured."
WAN optimization protects QoS
WAN optimization will play a strong role in protecting the performance of applications that might suddenly find themselves squeezed on the WAN by storage traffic. Storage replication can eat up all available bandwidth quickly.
"You're used to dealing with data networks where people are sending emails and other electronic attachments and not people sending out four gigabytes of data," Laliberte said. "You need to guarantee the performance of applications with some sort of QoS [quality of service]. "
The competition between storage and application traffic over a converged WAN was one of the drivers behind some recent strategic moves made by WAN optimization vendors, Cucchi said. For instance, Riverbed bought Mazu Networks last January. Mazu specializes in application performance analysis and reporting technology.
"It's a combination of gaining visibility into these WANs and being able to inspect that traffic and the quality of service for them," Cucchi said. "
Aaron , of Silver Peak said network engineers need network visibility when consolidating SAN replication on the WAN.
"You need to be able to track network behavior because that obviously impacts storage," he said. "Look at how much bandwidth is being utilized and how much exists, how much latency exists across the WAN and what impact that's having on your replication traffic.
Let us know what you think about the story; email: Shamus McGillicuddy, News Editor