Network vs. Hypervisor: The Battle for the Virtual Cloud Network is Very Real
Software-defined networking (SDN) is the next wave of innovation in networking, but achieving agility in the network that’s suitable for the cloud isn’t about hardware vs. software. What the industry needs to be talking about is “network-based” vs. “hypervisor-based” approaches to software-defined networks and network services that are most closely tied to applications. In other words, where should the networking intelligence lie?
The network-based camp supports that the intelligence needs to remain within the network core, where packets flow from the core out to the edge. Maintaining the intelligence here enables the network to do what it does best – keeping everything moving, balanced and secure.
The newly formed hypervisor-based camp contests that the intelligence –routing tables, firewalls, etc. - should reside on the edge of the data center in the VM, where the network meets the server.
Not only does this approach increase complexity and have potential budget and organizational ramifications, some services cannot be delivered within the hypervisor.
In this presentation, Dante Malagrino, CEO and co-founder of Embrane and former Cisco executive, will advocate for the network-based camp, so that attendees leave with a full understanding of the benefit of placing the intelligence in the network core of the data center. Malagrino is a seasoned industry speaker, having recently given presentations at leading events such as GigaOM Structure, Cloud Connect and Interop.
Malagrino will explain that the functionality and intelligence of virtualization is more efficient and better suited for the network, including the following points:
- Some network services cannot be implemented at the edge - Load balancing services, for example, must be implemented at the aggregation or core layers of the network to improve availability of end points (VMs) that are naturally located across multiple hosts.
- Controlling network services at the hypervisor layer is intrinsically more complex - The management density of hypervisor-based network services (one management point per hypervisor) is much higher than the management density of network-based network services (one management point).
- Deploying network services at the hypervisor layer increases resource fragmentation – When using the same type of resources (x86) for compute and network workloads, hypervisor-based implementations are less efficient than network-based ones.
- Hypervisor-based solutions are hypervisor specific and provide poor support for hybrid virtual/physical environments - Network-based solutions are designed to support both physical and virtual environments, as well as multiple hypervisors concurrently and without increasing the number of points of management.
- Network-based solutions scale better than hypervisor-based solutions - Hypervisor-based solutions will have to be over provisioned to account for future scaling needs, which exacerbate the resource fragmentation previously mentioned.
Building on the argument for the network-based camp, Malagrino will explain that the shift to networking intelligence in the hypervisor will not only be an architectural shift but could also cause a dramatic cultural shift in IT departments. Giving the systems and virtualization team, who is not well versed in networking, control of the network and compute could cause tension and may even require re-alignment within the IT organization as well adjustments in how budget and resources are directed.
Malagrino will conclude by discussing the notion that regardless of which camp you are in, companies and network professionals will ultimately have to expand their expertise to include programming skills, since programmability is one of the cornerstones of SDN. Software requires programmers to be able to code and script, which hasn’t been needed in the hardware world.
At the close of the presentation, Malagrino will open up the floor for discussion so that attendees can debate the two approaches.