Should server blade vendors be concerned with the software applications their customers are delivering? Conventional wisdom says no - just design the hardware to maximize performance on traditional axes: maximum compute power and maximum IO between CPUs, memory, network, and storage. But often surprises are found after hardware is in the field because designers don't anticipate how their customers intend to use the product. Trying to simply maximize hardware performance across all data paths can fail to address the key bottlenecks and issues that manifest themselves with emerging, rapidly evolving cloud computing use cases. 

As Ixia’s virtualization and cloud computing expert, I outlined strategies to assess system performance using leading cloud computing topologies at Server Design Summit in Santa Clara, California. Evaluating server architectures at the system level allows engineers to design hardware that will meet the needs of large cloud service providers and the endless use cases their customers deploy. 

Virtual switching performance is not up to par with hardware-based switching solutions. Inter-VM bridging performance can be improved by implementing switching functionality directly in hardware on compute blades. Cloud providers will be pushed by customers to make the network topology definition as flexible as compute resources currently are in clouds. Solutions to this challenge include using software defined networking (SDN). For more details, take a look at my Server Design Summit slides.