Abstract: Low-latency high-throughput scalable data caching is disclosed. For example, a data source is connected over a network to a load balancer server with data cache. A load balancer service and a data cache service execute on processors on the first load balancer server to receive, by the load balancer service, a request from a client device over the network. The load balancer service requests a data entry associated with the request from the first data cache service. The data cache service retrieves the first data entry from the first data cache, which stores a first plurality of data entries that is a subset of a second plurality of data entries stored in the data source. The load balancer service modifies the request with the data entry. The load balancer service sends a modified request to a plurality of receivers.

Abstract: Individualized connectivity based request handling is disclosed. For example, a content source is accessed by a client device and a load balancer executes on a processor to receive a first request based on the client device accessing the content source. A first session variable is set to a first value in a first session and a first latency to the client device is measured. A first plurality of target nodes is selected based on the first session variable. A first plurality of messages is sent to the first plurality of target nodes. A second request is received from the client device after the first session expires, starting a second session. The first session variable is set to a different second value in the second session. A second plurality of messages is sent to a second plurality of target nodes different from the first plurality of target nodes.

Abstract: Dynamic advertisement routing is disclosed. For example, a plurality of internet protocol (“IP”) addresses associated with respective plurality of target nodes is stored in a routing pool. Each IP address in the routing pool is pinged through each of first and second load balancer network interfaces. Network routes associated with target nodes are updated based on a first plurality of ping responses. Communications sessions are established with target nodes through respective network routes. IP addresses are pinged and respective latencies in a latency cache are updated based on a second plurality of ping responses. A first request directed to the plurality of target nodes is received and is determined to be sent to a first target node based on the latency cache forwarded to the first target node via the first network route.

Abstract: A video advertising system, methods, and apparatus are disclosed, which may include an advertising system, including an advertising server and advertising console, and a user device, which may cooperate to select video advertising campaigns and display video advertisements. In an example embodiment, a method includes playing a video on a display, overlaying markings on the video while the video plays, at a first time while the video plays, measuring a first refresh rate of the video, at a second time while the video plays, performing a first operation on the markings, measuring a second refresh rate of the video at the second time, and determining an estimated area of the video displayed on the display based on the first refresh rate and the second refresh rate. In an example embodiment, a prediction model is trained to output a viewability inference for a video.

Abstract: Dynamic advertisement routing is disclosed. For example, a plurality of internet protocol (“IP”) addresses associated with respective plurality of target nodes is stored in a routing pool. Each IP address in the routing pool is pinged through each of first and second load balancer network interfaces. Network routes associated with target nodes are updated based on a first plurality of ping responses. Communications sessions are established with target nodes through respective network routes. IP addresses are pinged and respective latencies in a latency cache are updated based on a second plurality of ping responses. A first request directed to the plurality of target nodes is received and is determined to be sent to a first target node based on the latency cache forwarded to the first target node via the first network route.