“…there is a real danger of spending more on storage unless the designers take a holistic view of the infrastructure when considering network performance.”

The feedback from IP EXPO 2009 confirmed that the virtualised approach to communications reaches into every corner of the network, from applications and operating systems to storage and files systems. A consequence is that a major challenge for the network professional is how to ensure optimum performance.

While vendors aspire to offer solutions that guarantee availability, ensure scalability and allow the whole infrastructure to be managed easily, a common gap in the marketing blurb is the problem that lies not in the individual network components, but in their interoperability.

Take a network comprising several dozen Virtual Machines (VMs) running in a network with generously apportioned CPU and memory to handle the workload according to the best available parameters. Yet the network fails to deliver satisfactory overall performance for the applications running on the VMs once density has risen above a certain threshold.

The problem, it turns out, lies in a bottleneck in devices outside the carefully managed server farm, notably in the storage arrays. These components perform to specific service levels and workloads, but these are not aligned with the workings of the server environment. The demands of the front-end server world have ended up working at cross-purposes with the back-end storage infrastructure, leading to poor overall system performance.

The Bigger Picture

Where resources allow, a natural response to avoid performance problems is to over-provision, such as spreading the workload of a storage array over an increasing number of disk drives. Would that life was so straightforward: other factors than sheer capacity affect the performance equation, such as the type of drive, its density, its speed and RAID level, the array architecture and caching characteristics.

While virtualisation can, indeed, save money through component consolidation -  particularly in the quantity of servers – there is a real danger of spending more on storage unless the designers take a holistic view of the infrastructure when considering network performance.

In the virtualized world, the physical components – such as actual servers and storage devices – are largely considered in the form of service layers. Performance therefore needs to be assessed across domains, involving combinations of network elements, rather than at the individual component level. The orchestra plays as one, with the violins and trumpets playing their roles in the context of the whole sound, rather than focusing on their contrasting individual timbre and volume.

This has the distinct advantage measuring overall system performance, recognising the various service layers, from servers to storage. The idea of adding a bit of speed here and a chuck of capacity there was invariably wasteful and expensive. Now there is a realistic alternative: the virtual network determined by accurate performance statistics.

Over the coming weeks and months we will begin to release audio and presentation content from our exhibitors and speakers which outline storage solutions and insight around virtualisation.