I/O innovations optimized for enterprise cloud

I and Sean Varley will be jointly presenting a session on ‘I/O innovations optimized for enterprise cloud’ at Intel Developer Forum (IDF). We are focusing on the I/O challenges in virtualization based enterprise cloud infrastructure and how current innovations, collaborations and technologies solve some of the challenges efficiently.

I have written in the past we have a view that evolution of virtualization has different phases to it. IT begins with basic consolidation, what we call virtualization 1.0, and then wants to extract more efficiency through flexible resource utilization and automation that we term as virtualization 2.0. The next phase to flexible resource management is deployment and management of scalable applications on a dynamic infrastructure, which we can relate to as enterprise cloud or virtualization 3.0.

In our view, the requirements of virtualization 2.0 makes virtualization 1.0 better and similarly the requirements of virtualization 3.0 makes 2.0 phase better. This means some of the challenges and solutions we discuss for enterprise cloud will make the IT datacenter today much efficient.

So what are the I/O challenges for enterprise cloud built on virtualization? There are many. Enterprise cloud model would mean being able to deploy the workload on available infrastructure (given that security and compliance needs are met) in a more flexible manner. This could lead to large scale consolidation given the new server capacity and performance. More VMs on a single server means more pressure on the I/O. So we need a balanced platform solution that maps the I/O capability to the CPU performance increases.

Flexible resource utilization should mean that the even the I/O hardware resources are flexible. However typical I/O architecture in a server is very rigid today. IT typically configures a virtualized server with a bunch of HBAs for storage I/O and eight or ten ports of Ethernet device for network traffic. Add to it separate cabling associated with each. Well, that means we cannot reallocate resources as needed in a flexible manner. So I/O hardware resource limits the extent of true flexibility. How to get around the challenge of rigidness in I/O architecture? And how do we reduce the complexity of the fabric and the power consumption? We perhaps need a unified converged high-speed I/O fabric.

Is it sufficient to have a converged fabric or do we need more? Of course we need more, what about the QoS and SLA of the I/O traffic? How about the scalability of the I/O fabric and security features to isolate the traffic between two VMs? All these are important as well in a multitenant enterprise environment.

In our session we explain how Intel in its products and through standards work has been targeting solutions for these challenges and delivering those to market with the ecosystem. To learn more make sure you attend the IDF session ECTS006. And for those who cannot attend, look for a blog from me and Sean post-IDF where we will succinctly define how we solve the challenges with Intel technology solutions.

RK Hiremane & Sean Varley