 If you are building latency sensitive data path applications and moving from hardware to containers, you can't make the transition to containers until the container environment can support a high throughput data path. Starling X enables a network appliance grade data path for containers and applications that need low latency and or high throughput. Starling X combines multi-CNI, enabling multiple interfaces for containerized network appliances. SRIOV-CNI, enabling direct access to virtual functions in a hardware interface. A VFIO driver, enabling access to SRIOV-VF interfaces in DPDK data paths. One gigabyte huge pages for containers enabling increased performance of memory access. ISOL CPUs and dedicated CPUs enabling predictable high performance via pinned and isolated cores. As well as an optional low latency kernel for optimizing low latency over high throughput for special case scenarios. While each of these interfaces is valuable on their own, Starling X has integrated and tested this technology so you don't have to. Together, they provide a powerful high speed low latency containerized data path and enable containerized cloud native network appliances. Starling X provides an optional low latency kernel for use cases demanding low latency. By providing a preemptible kernel and custom software and kernel tunings, Starling X can optimize low latency over high throughput. Here, we see some information about a data path performance test used to illustrate a high performance data path. Using an Ixia test set to generate and receive traffic using the test setup below, we run a container with multiple interfaces on Starling X. We modeled both the kernel based data path as well as the DPDK based data path as typically Starling X users will have some mix of both. With 10 gigabytes of traffic going in each direction and running the standard kernel, the kernel based data path achieves line rate at a packet size of 1,024 bytes and above. While the DPDK based data path achieves line rate at packet sizes as small as 128 bytes. In this view, we see the comparison between a high performance and a low latency cloud native data path. The test setup is the same except Starling X is now using a low latency kernel. This might apply when you need low latency in the signaling plane. For example, call setup with traffic that isn't sending a lot of data but needs very low latency. Here, you can see that the kernel based data path has an average latency that is 50 to 80% lower than seen with the standard kernel. Note in the table that this comes with a trade off of lower throughput than seen with the standard kernel. Note also that the user spaced DPDK data paths both throughput and latency are unaffected by the low latency kernel. These are sample results for this hardware setup. Your results may vary. In summary, Starling X provides a unique integration of a number of technologies which provides an environment to support network appliance grade data paths in a containerized environment with an ability to specialize with low latency or high throughput use cases.