In-Depth

Networking for the Cloud: 3 Vendor Views (Part 1, Emulex)

Three vendors, three opinions on what drives their solutions. First up is Brandon Hoff of Emulex, on the keys to the cloud and WANs that his company's solutions take advantage of to more easily move workloads among hybrid clouds.

• By Doug Barney
• 01/18/2012

First up is Brandon Hoff, director of product management for Emulex, a producer of high-speed networking devices for storage arrays and storage networks.

Redmond: What are the most important networking/WAN issues IT has to think about when considering the cloud?
Brandon Hoff, Director of Product Management, Emulex: WAN architectures have several considerations that are important in network design, such as the number of sites, location of sites, network bandwidth, redundancy and the number of remote employees. These already have been addressed in most enterprises. Also, the cloud provider offers connectivity as a service with a certain level of performance and availability at a given price, so service provider connectivity is already addressed.

With the cloud, the network and the WAN become an even more critical part of the infrastructure. The type of network and WAN connectivity depends on the type of cloud applications being accessed. On one end, SaaS [Software as a Service] offerings predominantly use Web browsers on client desktops. Additional bandwidth will be required, and it will scale based on the number of users and their location. HTTP over TCP provides a robust and ubiquitous infrastructure for connectivity and reliability. Additional bandwidth and load balancing may be beneficial and should be considered. Otherwise, SaaS is simple because it uses IP [Internet Protocol].

On the other hand, for hybrid cloud (Infrastructure as a Service, or IaaS) offerings where corporate workloads live in the cloud, things get more complex. First, virtual machines (VMs) are identified by their IP and MAC addresses, which makes VMs hard to move, but they have to be moved to get into the off-premises hybrid cloud. Second, while each VM workload uses IP, they also may use other types of Ethernet traffic that's required for the workload as well, so L2 connectivity is important. Third, service providers offer a combination of IP, MPLS, VPLS, metro Ethernet, dark fiber and so on. Fourth, add in a hybrid cloud provider where there are different switch vendors, different service provider connectivity solutions and so on. These all make hybrid cloud deployments complex and hard to do. The ability to onboard a VM to a cloud provider and move workloads between cloud providers, also known as VM mobility, is a hot topic and critical requirements for hybrid clouds to be successfully leveraged.

To reduce complexity and simply VM mobility, there's a new technique called overlay networking. Overlay networking uses encapsulation to tunnel network flows and isolate them in multi-tenant environments. To implement overlay networking, there are two proposed IETF [Internet Engineering Task Force] standards: NVGRE [Network Virtualization using Generic Routing Encapsulation] and VXLAN [Virtual Extensible Local Area Network]. NVGRE isn't a new standard, but a new way to use the tried and true Generic Routing Encapsulation (GRE) protocol between hypervisors. While all switches support both NVGRE and VXLAN protocols transparently because they're IP, GRE is already understood by switches and other network devices for administrators who need integrated functionality. In any case, overlay networking makes hybrid cloud deployments as simple to deploy as SaaS solutions because overlay networking leverages the ubiquity of IP for data flows over the WAN; therefore reducing complexity and improving flexibility and elasticity.

One last thing to plan for in hybrid cloud environments is large data transfers. Moving a VM and its workload to the cloud means moving all of its storage, which will require large amounts of bandwidth and take time. Plan for large data transfers and make sure your service provider can support your requirements when it's needed and that the cost is reasonable.

PaaS [Platform as a Service] uses IP protocols, so it scales and can be ubiquitously deployed across different environments. Performance will vary by application, but is purchased by the amount of data moved. Your application developers should have a good idea of the bandwidth required for each application and the bandwidth requirements for the WAN. Get their requirements and support them for your network architecture.

How does one evaluate the network/WAN's readiness for the cloud?
Focus on the services required for SaaS, PaaS and the hybrid cloud. First, are they all IP? Second, what is the bandwidth required for your SaaS applications? PaaS applications? And hybrid cloud workloads and VM mobility?

How do these issues differ depending on size of the enterprise?
The amount of issues due to cloud deployments doesn't vary much based on the size of the enterprise, it depends on the types of applications or products being used from the cloud. A big change should be expected for hybrid cloud deployments as virtual workloads move into the cloud.

How does your tool help solve these issues?
Emulex is working with Microsoft in providing NVGRE capabilities in Windows 8 and VXLAN capabilities in a future release of VMware. Emulex provides virtual networking for the hybrid cloud where VMs can be moved anywhere. Emulex also provides the raw bandwidth required for high-performance networking for cloud providers and enterprise customers.

Does the cost of additional bandwidth still make the cloud an economic advantage?
The economics of cloud computing -- and computing the TCO of a cloud offering -- has to include the incremental WAN costs when it's compared to an on-premises deployment. While the cloud can be used for day-to-day operations, the cloud also provides the ability to add capacity on demand to address seasonal workloads, or allows CIOs to delay the buildout of a new datacenter. For companies that ramp up compute workloads for the holidays, they can use the cloud to cover peak requirements so they don't have to over-buy hardware. For companies that are running out of compute capacity in their datacenters, moving workloads into the cloud until there's enough workloads to justify a new datacenter is also financially sound.

What will change in the future to mitigate some of these issues?
Key technologies that are accelerating the adoption of the hybrid cloud [include] overlay networking, which allows a virtual workload to move anywhere, as well as the move to 10Gb Ethernet and 40Gb Ethernet at the cloud provider and at the network core.

[Read Part 2 with CloudSleuth's Lloyd Bloom next week! -- Editors.]