ADQ Overview

Intel® Ethernet 800 Series Network Adapters feature Application Device Queues (ADQ), an Intel technology designed to prioritize application traffic to deliver the performance required for high-priority, network intensive workloads. By performing application-specific queuing and steering, ADQ can improve response time predictability, reduce latency, and improve throughput for critical applications.

ADQ improves performance in the following ways:

• Dedicated Resources = Increased Predictability

  ADQ leverages multiple Intel hardware technologies to steer, load balance, and isolate incoming application traffic to a set of dedicated application queues. By preventing other traffic from contending for resources with a chosen application's traffic, performance becomes more predictable and less prone to jitter.

• Efficient Packet Processing = Higher Throughput and Scalability

  ADQ ensures that all packets from a queue are processed in one thread and are for the same application, which enables polling for data to be more performant. For example, the application might accept a few packets, choosing to process them sooner, rather than spending more time accepting many packets, and so taking longer before any processing starts. Queuing attributes can be customized based on an application's traffic profile, for example, significantly reducing the number of interrupts per second.

• Customizable Traffic Shaping = Application Level QoS Control

  ADQ allows for dedicated application traffic queues, to have custom QoS (for example: rate limits) applied to them, allowing traffic to be shaped per application instead of per class or VM.

• Optimized Polling = Lower Latency

  Application Dependent Polling Method: ADQ provides hints to the application for each application thread to service incoming and outgoing traffic from a single device queue. By doing so, a queue is drained in order by one thread, minimizing context switching, cache misses, and sub-optimal thread scheduling.

  Application Independent Polling Method (ADQ 2.0): In this mode, an independent poller drains the packets from the device queues and pushes them to socket queues associated with application threads. This avoids the need for application threads to poll the device queues directly. As independent pollers can be used to poll multiple queues, a smaller number of pollers can service a large number of application threads and is more scalable. The poller threads minimize interrupts by polling the device queues, minimize context switches by pre-filling socket queues and avoiding the application threads from going through sleep/wakeup cycles. Although an application need not be modified to use the hints provided by ADQ to align the application threads to device queues in this mode, doing so can further improve the performance.