In the cloud datacenter, for the multi-tenant model, network resources should be fairly allocated among VDCs (virtual datacenters). Conventionally, the allocation of cloud network resources is on a best-effort basis, so the specific information of network resource allocation is unclear. Previous research has either aimed to provide minimum bandwidth guarantee, or focused on realizing work conservation according to the VM-to-VM (virtual machine to virtual machine) flow policy or per-source policy, or both policies. However, they failed to consider allocating redundant bandwidth among VDCs in a fair way. This paper presents a bandwidth that guarantees enforcement framework NXT-Freedom, and this framework allocates the network resources on the basis of per-VDC fairness, which can achieve work conservation. In order to guarantee per-VDC fair allocation, a hierarchical max–min fairness algorithm is put forward in this paper. In order to ensure that the framework can be applied to non-congestion-free network core and achieve scalability, NXT-Freedom decouples the computation of per-VDC allocation from the execution of allocation, but it brings some CPU overheads resulting from bandwidth enforcement. We observe that there is no need to enforce the non-blocking virtual network. Leveraging this observation, we distinguish the virtual network type of VDC to eliminate part of the CPU overheads. The evaluation results of a prototype prove that NXT-Freedom can achieve the isolation of per-VDC performance, which also shows fast adaption to flow variation in cloud datacenter.
Fine-grained load balancing with traffic-aware rerouting in datacenter networks