Scalable cloud computing

A method for segmenting traffic from a plurality of customer accounts within a scalable cloud infrastructure, the method comprising:

• allocating one or more units of virtual resources to each of a plurality of customer accounts, each of the one or more units of virtual resources corresponding to physical resources partitioned into a plurality of physically isolated zones within the scalable cloud infrastructure, the plurality of physical resources comprising a plurality of physical compute, storage, and networking resources within a plurality of data centers distributed across a network, wherein the plurality of physical resources are shared at least in part between two or more of the plurality of customer accounts, and wherein for each customer account, the physical resources are selected from one or more of a plurality of weakly connected pods within one of the plurality of zones, each zone including one or more zone-level layer-3 network switches providing interconnections among the pods of the zone and providing connections to other zones, each of the plurality of pods comprising a strongly connected discrete set of the plurality of physical resources within the zone, the physical resources in each pod including one or more compute servers, one or more storage servers, and one or more routing servers connected together by one or more pod-level layer-2 network switches;
• and
• for the physical resources that are shared at least in part between two or more of the plurality of customer accounts, segmenting traffic from the one or more units of virtual resources allocated to a first of the two or more of the plurality of customer accounts from the one or more units of virtual resources allocated to a second of the two or more of the plurality of customer accounts;
• wherein the compute servers of a pod execute a multitenant hypervisor, and the storage servers of the pod present virtual disks to the compute servers, the virtual disks being visible to the multitenant hypervisor as networked file systems or as block devices and being presented as locally attached storage to virtual machines of the compute servers.

A system for a scalable cloud infrastructure, the system comprising:

• a plurality of physical resources partitioned into a plurality of pods within a plurality of zones, wherein:
• the physical resources comprise physical compute, storage, and networking resources within a plurality of data centers distributed across a network;
• each zone comprises a weakly connected subset of the plurality of pods and is physically isolated from other zones in the plurality, and one or more zone-level layer-3 network switches providing interconnections among the pods of the zone and providing connections to other zones;
• and
• each pod comprises a strongly connected discrete set of the plurality of physical resources in the zone, the physical resources in each pod including one or more compute servers, one or more storage servers, and one or more routing servers connected together by one or more pod level layer-2 network switches;
• at least one data network from the distributed networking resources, the at least one data network connecting, and configured for routing traffic to, the plurality of pods from at least one customer;
• at least one management server cluster comprising a plurality of management servers;
• at least one management network from the distributed networking resources, the at least one management network connecting one of the plurality of management servers to one of the plurality of zones, and connecting the compute and storage resources partitioned within the pod;
• a plurality of units of virtual resources by apportioning selected of the physical resources within one or more of the plurality of pods to each of the plurality of units, wherein the apportioned physical resources may be shared between two or more of the plurality of units of virtual resources;
• and
• wherein each management server is configured for allocating one or more of the plurality of units of virtual resources to an account associated with the at least one customer;
• and
• wherein the compute servers of a pod execute a multitenant hypervisor, and the storage servers of the pod present virtual disks to the compute servers, the virtual disks being visible to the multitenant hypervisor as networked file systems or as block devices and being presented as locally attached storage to virtual machines of the compute servers.