Archive for August 17th, 2009

“Future is bright: Cloud of Clouds”

Monday, August 17th, 2009 by Antonio Manzalini

 

Title of this post is a quotation of Lew Tucker (Vice President and CTO of Cloud Computing at Sun Microsystems, Inc). He has made an interesting speech on the evolution of computing as it moves, through virtualization and autonomic, into a self-sufficient cloud.

 

Please see a glimpse of the speech at:

 

http://virtualization.sys-con.com/node/1072553

 

Resources virtualization provides a level of indirection and it allows resources partitioning: as a matter of fact, the former increases the flexibility, the latter results in a more effective resource sharing. Moreover different administrators can manage different virtualized resources on the same hardware, enabling this new biz models. Of course virtualization is not new; in the last few years it has become a hot topic that has seen the development of commercial solutions such as Xen and VMware.

 

Extending this model to the network resources (for example routers and switches), in general, provides flexibility, promotes diversity, and promises security and increased manageability by allowing multiple heterogeneous network architectures to cohabit on a shared physical network. Existing technologies provide already coarse-grained link virtualization. Virtual Private Network (VPN) connecting multiple distributed sites through tunnels over shared or public networks (e.g. L3 VPN based on IP and MPLS, L2 VPN based on Ethernet, L1 VPN based on circuit-switching and VPN at the application layer). An Overlay Network is again another form of network virtualization which is typically implemented in the application layer, even if various implementations at lower layers of the network stack do exist. PlanetLab represent an interesting example: Users (mainly researchers as PlanetLab is a scientific test-bed) can easily create and manage customized environments to design and evaluate new architectures, protocols, and algorithms. Also peer-to-peer networks (e.g. BitTorrent) can also be seen as examples of virtual networks. Vendors such as Cisco and Juniper offer router virtualization. Virtual Machine Monitor (VMM) features the ability of a physical router to host on or several virtual machines on its hardware. Nevertheless, all these examples suffer from a lack of node performances and programmability, sufficient isolation, cross-layer and cross-domain interoperability, optimization of physical resources, etc.

 

Now, let’s think about an augmented model of network virtualization: imagine a Cloud Networking open environment spanning across interoperable network domains, capable of supporting multiple coexisting, programmable and cooperative virtual networks.  These virtual networks are obviously composed by virtual nodes connected together by a set of highly dynamic virtual links. Virtual nodes can share, in an optimized way, common pools of physical resources across the various network domains. Virtual nodes can either be a virtual host or a virtual router. A virtual host acts as a packet source or a sink; a virtual router (i.e. forwards packets according to the protocols of that virtual network). Not only routers or switches are virtualized, but also smart-phone, other Users’ device (laptop, PDA, etc) and anything of Future Internet. Virtual nodes can migrate across hardware physical domains to exploit at best traffic engineering policies, to augment execution capabilities (e.g. cloning) or simple to optimize physical resource allocation.

 

This open virtual environment can become an open ecosystem capable of creating new and variegated biz opportunities; moreover virtualization and autonomic decentralized self-control of resources can determine network costs dropping quickly.

This scenario implies many challenges, still open. Ability to control and move these virtual nodes creates a need for an intelligent architecture that can adapt to changing demands and conditions by autonomically tuning performance, isolation, migration, and sliver resizing. Among others, there is also a need to find innovative solutions for developing cross-domain virtual control planes, high performance virtual forwarding planes, and to make the forwarding planes re-programmable: with this capability, one can think about self-organizing forwarding paths where they adapt dynamically to the incoming traffic or applications requests (even for new forwarding path experiments without affecting the operation of the existing network, for example see OpenFlow).

In conclusion, innovative solutions for network virtualization already provide a powerful approach to run multiple customized networks at the same time over a shared substrate of physical resources. Advances in this direction, coupled with autonomic self-management and control, can pave the way to Cloud Networking. This seems showing a lot of promises, from cost savings, to enabling new biz model, from a complete decoupling network vs services to overcoming Internet tussles and ossification.