Transparent load-balancing for cloud computing services

Systems and techniques are described for virtual machine (VM) load-balancing. A described technique includes running VMs that communicate via a virtual network, the VMs being configured to handle requests associated with at least a destination address on a public network and operating gateways that provide connectivity between the virtual network and the public network, and use a consistent mapping operation to perform load-balancing among the VMs. A gateway can obtain VM health information; determine based on the VM health information a group of healthy VMs; receive an incoming packet that includes a source address and the destination address; select a handling VM based on an output of the consistent mapping operation given the source address and the group of healthy VMs; forward the incoming packet to the handling VM via the virtual network; and create an entry to forward one or more additional incoming packets to the handling VM.