Fabric Computing

Fabric computing, or unified computing, is the principle driving state-of-the-art next generation data centers and networks. All resources are linked together in a common architecture that can be virtualized. In other words, the compute and storage platform is architecturally "unified" with the network and the virtualization platform, making no distinction between the network and the edge devices connected to it. Fabric computing eliminates manual integration, in favor of an integrated architecture, thereby breaking down the silos of compute, storage, and network resources.

 

1.Architecture

Fabric computing may be built into a particular vendor's computing equipment, such as Cisco's Unified Computing System. Alternatively, this unification may be realized through the use of middleware linking the different systems together.Ultimately, both approaches need to be supported by the GreenStar Network; however, due to the cost of specialized hardware and the need to utilize existing equipment, only the middleware approach will be taken within this project.

Figure 1: Unified Computing and Virtual Infrastructures

Figure 1 shows how fabric computing facilitates virtualization & delivery of infrastructure. A key bility of fabric computing is to easily virtualize any part of this fabric – known as “slicing”. For this project, the optical network elements, servers, routers and disc arrays will be unified through the use of various middleware solutions and tools. In order to achieve the required level of mobility, software instances of will be used. While software does not offer the same performance as the hardware assisted counterpart, it is ideal for understanding how the hardware counterparts must behave, and will be sufficient for the project’s GSN implementation.

 

2.Tools

To establish this fabric computing environment at each GSN node, the tools listed below will be used. The IaaS framework, an open source framework for resource description and exchange, will be used to wrap the different cloud solutions, tools and service offerings at the nodes.

  • XORP and Quagga will be the supported software based routers and will be running on dedicated virtual machines.
  • Xen and KVM will be the supported hypervisors, since most cloud computing middleware is based upon these. Puppet will be used for contextualization.
  • VDE and Linux Kernel will be used for Ethernet Management.
  • iSCSI and LUN, as well as network file systems (NFS, GFS, etc.) will be used.
  • Linux will be supported operating system used by hosts and guests.

As additional leverage to the GSN, the Netvirt project will be providing an environment to provide hardware assisted processing for the software-based routers, using the Tilera TileEmpower processing platform.

Figure 2: IaaS Framework's Architecture

The IaaS Framework will be used as the integration middleware, it's architecture depicted above makes it suitable to represent any device or service as resources. Efforts are currently underway to create wrapper around OpenNebula and that will provide the basis on which these deliverables will be done. Moreover, while these tools offer the core technologies on which the GSN fabric will be built, management middleware will be needed. The following middleware solutions, commercial(indicated by $) and opensource, will be used:

  • OpenNebula, Eucalyptus and Incus($) will manage servers and virtual machines.
  • Ether($) will control VDE, Linux Kernel and the Allied Telesis switches.
  • MANTICORE and Pike($) will control routers and virtual routers.
  • Walrus will control the storage system.

These diverse middleware solutions will be integrated by wrapping them, and translating the resources models, to conform to IaaS Framework resources.