Ask a dozen people what cloud computing is and you're bound to get a dozen different answers. To some, it's all about service levels. To others, data migration and load balancing. Still others like the idea of unlimited resources and anywhere/anytime applications.
Most people agree, however, that the underlying technology that makes it all possible is virtualization. And while that's true, the level of virtualization you need to join the cloud is wholly dependent on the kind of I/O infrastructure you put in place.
10 GbE is likely to be the networking protocol of choice as the clouds roll in, although some of you higher-end folks are probably just as likely to head into 20 G Infiniband and up. But the question then becomes, will your underlying virtual/cloud platform have the chops to take advantage of such high-speed networking?
Already, the leading virtualization vendors are starting to stress their I/O capabilities as they transition over to cloud formats. VMware is the latest and clearest example with the new vSPhere 4 platform released earlier this week. While the system is loaded with tools and techniques designed to bring cloud provisioning and management under one roof, one upgrade that's hard to overlook is the 200,000 IOPS it can handle, twice as many as the Virtual Infrastructure 3 platform it's designed to replace.
And that's only scratching the surface, according to this report from the VMware and EMC performance teams on Virtual Geek. The groups are reporting more than 325,000 single ESX hosts and counting and, in combination with the CX4 storage systems as few as 30 flash drives can replace more than 2,000 Fibre Channel drives. Expect full details at EMC World next month in Orlando.
EMC is also targeting virtual and cloud I/O bottlenecks with its new PowerPath/VE software, an update of the company's multipath system designed to shift workloads in networked environments. Unlike earlier implementations, however, the PowerPath system does away with monitoring and manual rebalancing of data loads by dynamically adapting to changing conditions. It also forges a strong link between vSphere and EMC's V-Max storage system.
Despite all the whiz-bang technology in the vSPhere, however, its performance can still be hampered by some familiar problems. One of them is disk fragmentation, according to this article by Bruce Boyers, who's written a fair amount on behalf of Diskeeper over the years. Nonetheless, he correctly points out that the coming and going of VMs in a virtual environment can wreak havoc on the orderly storage of information. It won't matter a hill of beans to have the fastest, most dynamic network on the planet if the disks themselves get bogged down searching for disparate bits of data.
So while it's true that the cloud is about services and resources and data flexibility, underneath it all, it's about infrastructure. Everything you need to get the job done will be available, but the data has to come together quickly.