Virtual Machine Device Queues

Boost network I/O to meet the demands of consolidated virtual workloads. Multiple Virtual Machines (VMs) introduce heavy traffic-management demands that servers must meet in order to ensure scalability and get the full value of virtualization.

Virtual Machine Device Queues (VMDq), a component of Intel® Virtualization Technology for Connectivity, optimizes the processing of VM data traffic to improve CPU utilization and bandwidth.

VMDq enhances virtualized traffic management

As the number of VMs on a server increases, so does the amount and complexity of traffic. VMDq manages the VMs' data traffic efficiently in order to reduce the I/O bottleneck in the system:

Throughput: Provides an alternative to VMM-based packet sorting, to ease throughput limitations
Scalability: Creates parallel data I/O paths in the network I/O silicon to avoid performance degradation as the number of VMs increases
Capacity: Liberates CPU cycles otherwise consumed by packet sorting, making them available to applications

These advances promise to increase server-consolidation ratios, adding to the cost savings associated with virtualization solutions.

Solutions for virtualized I/O challenges

VMDq offloads the sorting burden from the VMM to the network controller, to accelerate network I/O throughput.

Together, these capabilities of VMDq improve the robustness of network connectivity to provide better traffic management capabilities to the VM data traffic:

Hardware-based prioritization and queuing reduces the burden on the VMM by allocating individual VMs' data to respective hardware queues to improve overall efficiency.
Additional data queues make the data path to the network interface parallel rather than the traditional serial, per-packet model, allowing VMs to more efficiently share network ports.
Packet sorting by the network interface hardware for incoming data removes that burden from the VMM software, avoiding I/O processing bottlenecks.
"Round-robin queue servicing by the network interface hardware improves transmit fairness and avoids head-of-line blocking among VMs, better enabling bandwidth efficiency and quality of service.