Extreme's Data Center Play

Extreme Networks has taken another step in articulating its data center vision with two new announcements to their data center product line. The first addresses applying policies to network traffic to and from the VM, and having those policies follow the VM during moves and disaster recovery events. The second is Extreme's stake in the ground supporting hypervisor switch by-pass that allows the hardware switch to move frames from VM to VM. Both announcements address architecture design requirements needed to support a virtualized data center. Extreme has to come up with something unique to compete with Cisco's Nexus architecture, HP's newly acquired H3C switch line, as well as Juniper and Brocade in the data center.

XOS Network Virtualization (XNV) is Extreme's foray into supporting virtualization in the data center. XNV, the company claims, provides visibility, control and automation to a virtual infrastructure by tracking virtual machines and applying port policies as VMs move from hypervisor to hypervisor. Not only do port policies ensure that virtual machines are properly supported on the network, but they can also be used to stop virtual server sprawl by restricting unauthorized VMs from coming on-line.

Port policies define the characteristics that are applied to switch ports. In a data center where there is a 1:1 relationship of a server to a port, administrators can apply QoS and access controls at the access port that grooms ingress traffic from the server. Data centers relying on physical servers are largely static and don't move. In a virtualized data center, not only can there be multiple servers on a physical port, but each virtual server can move from physical server to physical server. The servers' port profiles should move with them. Virtual port profiles apply QoS and access controls to virtual servers and should move with the server.

When a new VM is provisioned and assigned a port profile, XNV performs the server identification and assigning port profiles automatically. XNV was designed to be automated. Working with Extreme's Epicenter management system, the switches will detect VM moves and will apply the port profile wherever it lands, even if that is on a different switch. Many vendors have virtual port profile capabilities, so this is a must have feature. XNV works with both VMware's vCenter and Microsoft Hyper-V.

Many infrastructure vendors profess a move to flatten the physical network and remove multiple switch tiers. For example, data center switching often uses a top of rack switch to connect servers to the network, then top of rack switches connect to end or row switches or directly to core switches. The number of tiers largely depends on the amount of physical aggregation required to get hosts onto the network. Direct Attach is Extreme's attempt to remove one or more tiers, reducing capital and operational costs with additional tiers and improving network performance by reducing hop-by-hop latency.Key to Extreme's Direct Attach is the 96-port fan out modules in the BlackDiamond chassis that can support high port counts when home running servers to a core switch. That's a lot of cable to run from rack to switch and may not be a viable option. It's one of the reasons why top-of-rack and end-of-row switches are used: the cable trays fill quickly as you approach the core.

The second necessary feature is Extreme's adoption of using a physical switch to move traffic between hosts, even within the same hypervisor. The idea, also promoted by HP and others, is to bypass the hypervisor soft-switch, which has limited functionality, and use the existing switch infrastructure. The argument is that doing so makes management better and simplifies the overall network topology. This is in contrast to Cisco's approach with the Nexus 1000V, which replaces the virtual switch in VMwares hypervisor, and Arista's approach of managing the soft-switch using their switch operating system.

Using the bypass method leverages the existing switch infrastructure but also has the potential to create hairpins where traffic between two or more VMs on the same hypervisor exits the server and then does a hairpin turn to come back to the same hypervisor. Work within the IEEE on 802.1 Qbg Edge Virtual Bridging  has just begun; the working group was approved in December of 2009. Given the typical working time within the IEEE, we can expect to see standards in 2-3 years. Extreme's Direct Attach is based on early work within 802.1Qbg, but it's a stretch to say it's standards-based. Direct Attach currently only works with the open source Kernel Virtual Machine (KVM) on RedHat, but Extreme plans support for other hypervisors in the future.

Whether virtual switching is going to be good enough or using Virtual Edge Bridging is going to be the way to go is yet undecided. What is clear that the network edge in the data center will have to increase capacity beyond Gb or multi-Gb connections, and that grooming traffic on a per VM basis will be necessary. Extreme, like other switching vendors, is getting their data center switching products ready.