Your cloud platform, the basic functionality you need to deliver cloud services, fundamentally relies on three key building blocks, your servers, your storage and your network. To allow clouds to respond rapidly to the demand of the business users, the first two have been virtualized, creating an abstraction layer between the actual infrastructure and the cloud environment. Automation allows for provisioning, management, use and de-provisioning of server and storage resources. Self-provisioning gives the user the capability to initiate that process himself. However, these resources need a network to integrate. And that is where in many datacenter the issue currently relies.
If you are looking at provisioning a virtual machine, frankly, you’re perfectly fine with just having server and storage virtualization, but if you want to provision a complex service requiring multiple VM’s, one or several databases etc. you need to set-up secure connections between the different elements. That is where the network actually plays an important role. In particular, if you are interested in being able to move your VM’s around to balance load in your datacenter, to account for high availability etc., having an agile network configuration capability is key. You want immediate action, so you’re not ready to wait for network operators to set-up all devices appropriately.
The need for a new network architecture
As the Open Network Foundation (ONF) points out in a well-documented white paper, titled “Software-Defined Networking: The New Norm for Networks”, most conventional networks are hierarchical. They have been built with layers of Ethernet switches arranged in a tree structure. The design evolved that way to address the needs of client-server computing where the bulk of the traffic was between the client and the back-end server (north-south traffic). In a cloud environment things are different. The majority of the traffic is between the virtual machines, and with the databases. To quote ONF, we are replacing traditional “north-south” traffic with “east-west” machine-to-machine communication. So messages keep hopping up and down the layers of switches making communication inefficient. This can be corrected by building new, flatter network architectures that are “context aware” and “application fluent”.
Combine this change in network traffic resulting with the need to support BYOD devices (increasing the need to protect corporate data and intellectual property while meeting compliance) and the integration of both private and public cloud services; you will very quickly realize we also need to look at network management differently. Add to this the “big data” phenomena and you will understand the need for larger bandwidth in the datacenter.
As services are provisioned and de-provisioned all the time, the demand on the network is constantly evolving. The regular introduction of new services and the obsolescence of others changes demand patterns regularly. Network operators can no longer coop with this extra workload associated with the manual reconfiguration of the switches and require automation and virtualization of the network.
Software Defined Networks
Not only is the traditional datacenter network hierarchical, but its control software is distributed across the actual network devices, in other words the switches. The concept of Software Defined Networking (known as SDN) is based around the decoupling of the network control and the actual traffic forwarding. If we perform control using software implemented on standard servers, we abstract the underlying architecture and are able to treat the network as a logical or virtual entity. The application can now request the networking required using standard API’s. The SDN control software will orchestrate the network resources and send commands to the actual devices to set-up the appropriate configurations. Network policy management, security, traffic engineering, access control etc. can all be addressed through a set of configurable applications in the SDN controller. An entire network can now be managed through intelligent orchestration and provisioning.
Need for standards
In a traditional environment, companies choose one brand of network devices that addressed all of their needs to ensure compatibility, facilitate the training of the network operators (hoping the brand used the same tools and configuration languages for all their devices) and improve responsiveness of the networking team. But as use cases multiply it becomes difficult to address all your needs with one brand. So, ideally you would like to avoid lock-in, giving you the capability to make the best choice for a particular need. SDN might help you in there as your network control interactions are now going through one software environment. But that environment has to deal with all the devices out there. In 2008, a group of academia have proposed a standard protocol for network devices, called OpenFlow. As pointed out by ONF, using an OpenFlow-based SDN architecture provides following benefits:
- A centralized control over multi-vendor environments
- Reduced complexity through automation, and I would add to that, reduced human errors. If I remember correctly, Amazon’s April 2011 outage was due to human errors in the reconfiguration of network switches.
- Increased network reliability and security
- More granular network control
ONF actually adds two other benefits I’d like to discuss a little more. Using SDN, networks can be reprogrammed in a matter of minutes to meet specific business and user needs. This can potentially change the user experience completely and allows for innovation in services delivered and capabilities provided. So including SDN in the cloud platform is a formidable way to open the door for innovation and better user experience.
What is HP doing in this space?
As in many other areas, HP is a proponent of OpenFlow as we believe in the importance of standards. HP has been developing three complementary components to provide you the flexibility needed to operate modern cloud environments:
- HP’s FlexNetwork Architecture simplifies and flattens the architecture across the datacenter (FlexFabric), the campus (FlexCampus) and the branch (FlexBranch). FlexManagement delivers a single-pane-of-glass for managing the architecture. Its core platform is the Intelligent Management Center (IMC), an SOA platform with modules delivering full fault, configuration, accounting, performance and security management. It provides centralized control and orchestration of the network.
- To virtualize this network architecture, HP has developed the Intelligent Resilient Framework (IRF), a switch virtualization technology, grouping the switches together under a single IP address, allowing them to be managed as a single switch. The entire network is now virtualized by providing a single control plane across the end-to-end network. This allows the entire network to be orchestrated using policies instead of managing device-by-device.
- Virtual Applications Networks (VAN) complements the first two and provides cloud functionality that speeds application deployment, simplifies management and supports network service level agreements across the HP FlexNetwork Architecture. VAN connect users to an application by designing and implementing the networks to meet the requirement of application classes, such as voice, real-time transaction processing, or video over a secure, shared infrastructure. VAN supports multitenancy. So, in a nutshell, network provisioning is separated from device management through the use of connection profile templates. VAN manager runs on IMC and contains three components, a designer, a policy engine and a VMware plug-in. Other plug-ins for Hyper-V, Xen and KVM are planned.
As pointed out by ESG, HP is putting a strong vision forward to enable networks for the cloud computing era. We do this on open standards and to prove how serious we are, we have announced that 16 of our switches already support OpenFlow. So, to implement your cloud environment, you may want to look at all three areas of servers, storage and networking for virtualization. And if you are a telco service provider, you may want to complement this with the joint ALU-HP offering allowing you to manage bandwidth and latency in your core network and provide your customers a unique user experience.
If you are interested in reading more on the subject:
- OpenFlow: Enabling Innovation in Campus Networks, March 2008
- Software-Defined Networking: The Norm for Networks, ONF, April 2012
- HP Virtual Application Networks, ESG White Paper, April 2012
- Orchestrate virtual application networks accelerate the move to cloud, HP business white paper, Apri...
- Reducing Network Complexity, boosting network performance with IRF technology, September 2010