While deskstop virtualization is truly becoming a reality and hot topic in the market, IOPS has become the major concern when designing and thinking to scale a virtual desktop infrastructure.
As opposed to server virtualization solution, desktops significantly increase the VMs density, surpassing the well known limits of previous virtualization solutions being IOPS the big matter, as biggest impact used to be the storage infrastructure cost to sustain a virtual desktop infrastructure.
IOPS are still a hot topic, but would it in the future. Storage solutions are quickly adapting to provide a lower cost IO effective solutions. From caching technologies to SSD drives and storage tiered architectures, technology is caching up on IOPS limitations for VDI deployments.
Currently brokers communicate with the infrastructure manager, being pool master on XenServer, vCenter for Vmware or VMM for Hyper-V. This guys weren’t originally designed for desktops but for servers management. Think about the scalability numbers/limits. The higher load I’ve seen has been with XenServer reaching out 3000 desktops, but that still low for large scale deployments.
Moving into a distributed architecture. What if hypervisors were not pool or cluster dependent, and the broker will take care of the VMs management. I’m sure there are some challenges with that, mainly API and security risks. However, with hypervisors hosting desktops on local storage when IOPS won´t the the problem anymore and with an architecture which will allow vm’s disks to travel across server’s for vm creation, shared storage wouldn’t stop being the TCO/ROI stopper with a far more scalable solution.
XenDesktop/XenServer Intellicache will definitely fit there, being a huge step forward on making ROI affordable for customers. There is still other technologies which many need to support this trend reducing the IOPS limitation on the hypervisor level. But this is certainly the way to go.
So will IOPS matter? I believe it will stop being a cost concern within the next 2 years timeframe.