This is a battle in which everyone wins. However - the information released is less valuable without information on how the systems are used. For example, it w

This is a battle in which everyone wins. However - the information released is less valuable without information on how the systems are used. For example, it would be great to get an idea of how Facebook is deploying the FusionIO cards they buy - and what the architecture is for serving up different components of its site. There was no mention of storage - we basically saw compute nodes and the power infrastructure. Are those nodes hosting multiple VMs? How are they connecting to storage. Where is the Cache Layer - It looks like, via the statement they can send a reboot command via the network - that they dont care about OOB management - which really makes sense when their entire datacenter is purpose built for a single site. So we see with both google and facebook a single application scaled to a global level. We dont see how these architectures can be modularized to work for smaller apps or multiple separate tenants in a single colo designed around this architecture. There is probably some good networking designs we can employ in a commercial data center to achieve compartmentalization of services -- although with the modern cloud - people area lot less sensitive to collocating their apps/data than they were just a few years ago. One of the challenges large traffic cloud hosted services are seeing is limitations in not owning the HW architecture, such as Reddit where they dont have the advantage of designing, say, an SSD front-end cache layer.