THE proven cost savings and high return on investment from virtualisation and cloud computing in the datacentre have captured the imagination of many CFOs and CIOs.
In most cases, organisations can reduce datacentre costs by moving more applications onto fewer servers and can reduce software licensing fees and administrative resources by migrating to a software-as-a-service (SaaS) model.
When they add in faster deployment times for new applications and services while ensuring 24×7 reliability, organisations have a compelling business case for virtualisation and cloud computing in the datacentre.
At some point in the future, this type of highly virtualised, services-on-demand delivery model enabled through cloud computing will be the IT gold standard.
In terms of the physical network infrastructure, virtualisation and cloud computing complement each other in the following ways:
1. Cloud computing services can reside on top of virtual datacentres. Virtualisation can support cloud architectures;
2. Cloud computing software can orchestrate virtual server deployments. Cloud management software can control virtualisation services;
3. Cloud computing adds another virtualisation layer between end users and the entire IT environment, which enables a pay-per-use model; and
4. Both virtualisation and cloud computing require robust physical network infrastructures. They rely heavily on the network and require new ways of thinking about network architecture and design.
Before this transition occurs, however, the underlying network architecture must provide greater availability, performance, and security while simplifying datacentre operations.
A complete revolution in the datacentre, with its associated costs, disruptions, and time requirements, is unnecessary in almost every case. Instead, the transition to a highly virtualised datacentre can be smooth, non-disruptive, and full of business advantages.
As organisations add Virtual Machines (VMs) and shift to a cloud-based operations model, the network must be hardened against failures that disrupt traffic yet adaptable and flexible enough to support new business requirements.
Many organisations are considering the option of network convergence, which combines SAN traffic and IP network traffic on the same physical network as a way to reduce the amount of cabling in server racks and the number of top-of-rack switches.
However, there are important tradeoffs that will limit adoption of this option, because the network architecture has to support both network convergence and non-converged traffic. High-availability and high-performance applications that are not well suited for convergence will remain on separate networks. Network latency, change control processes, regulatory compliance, and security are other factors that might require IP and storage traffic to remain separated.
When and how far to converge the IP and Fibre Channel traffic - or not to converge - is a decision that should be made in the context of the unique requirements of each organisation.
Organisations can implement a virtualisation and private cloud computing architecture (converged or not converged) that makes sense for their particular business requirements.
The initial step is to decide how the datacentre should look in the future by determining the extent of LAN/SAN convergence, the number of layers within each network, the number of switching tiers in each layer, and the management model for virtualisation and cloud computing services.
Start by addressing the following questions:
a. Will the final network have physical access layer switches in each rack, or will that function reside inside virtual servers?;
b. What level of oversubscription is acceptable at each stage of the network, given the target server consolidation ratio?;
c. How much bandwidth would each VM receive if they all tried to communicate simultaneously?;
d. Will there be an aggregation layer on the LAN, or will large virtual servers connect directly to a high-port-count collapsed access/aggregation layer?;
e. Will there be a single orchestration tool and hypervisor vendor, or will there be multiple solutions for specific applications and departments?; and
f. Will there be continued use of any existing equipment in that design? And if not, why not?
The key to achieving "revolution through evolution" in datacentre networking is to move toward a target design along a well-planned path and to use incremental steps to control risk.
(Sean Ong is country manager for Malaysia at Brocade Communications Systems, a networking solutions vendor.)