Abstract: An online system stores user profiles of users performing past user interactions with content items. The system receives a new content item and extracts a new feature vector from an image in the new content item using image analysis. The system generates, by a machine learning model, a demographic criteria vector based on the new feature vector. The machine learning model is configured based on the user profiles of the users performing the past user interactions with the plurality of content items to receive a feature vector for a content item and generate a demographic criteria vector based on the feature vector. The demographic criteria vector indicates a likelihood of a user with a user profile matching the demographic criteria vector interacting with the content item exceeding a threshold. The system sends the demographic criteria vector to a content provider for targeting the new content item.

Abstract: An online system receives related content items from a content provider for delivery over a period of time. The system determines a rate of delivery of each related content item to users of the online system. The system delivers the related content items based on the determined rates of delivery. The system monitors a rate of client interactions with the related content items. The system adjusts the rate of delivery of each related content item based on the rate of client interactions associated with the related content item, such that higher performing related content is delivered more than lower performing related content. If the rate of delivery of the related content item decreases below a threshold, the system sets the adjusted rate of delivery of the related content item to the threshold. The system delivers the related content items to client devices based on the adjusted rates of delivery.

Abstract: An online system receives related content items from a content provider for delivery over a period of time. The system determines a rate of delivery of each related content item to users of the online system. The system delivers the related content items based on the determined rates of delivery. The system monitors a rate of client interactions with the related content items. The system adjusts the rate of delivery of each related content item based on the rate of client interactions associated with the related content item, such that higher performing related content is delivered more than lower performing related content. If the rate of delivery of the related content item decreases below a threshold, the system sets the adjusted rate of delivery of the related content item to the threshold. The system delivers the related content items to client devices based on the adjusted rates of delivery.

Abstract: An online system stores user profiles of users performing past user interactions with content items. The system receives a new content item and extracts a new feature vector from an image in the new content item using image analysis. The system generates, by a machine learning model, a demographic criteria vector based on the new feature vector. The machine learning model is configured based on the user profiles of the users performing the past user interactions with the plurality of content items to receive a feature vector for a content item and generate a demographic criteria vector based on the feature vector. The demographic criteria vector indicates a likelihood of a user with a user profile matching the demographic criteria vector interacting with the content item exceeding a threshold. The system sends the demographic criteria vector to a content provider for targeting the new content item.

Abstract: A method and system in data processing system are disclosed for the dynamic scheduling of a plurality of requests to access a disk. Each of the requests is associated with a location on the said disk which each of the requests is attempting to access. A scan queue is established for storing the plurality of requests. The plurality of requests are processed in a sequential order. The sequential order is determined utilizing the location on the disk being accessed by each of the requests. Upon one of the stored requests being urgent, the urgent request is processed. The urgent request is associated with a first location on said disk. Processing of the requests then continues in a second sequential order. The second sequential order is determined utilizing the first location. The next request to be processed is one of the requests which is associated with a physically closest location on the disk.

Abstract: A method and system provide end-to-end flow control between a sender and a receiver attached to a network. The data is divided into a plurality of packets. A priority hierarchy of a plurality of priority levels is established. Each packet of data is associated with one of the priority levels. Transmission of the plurality of packets is initiated from a transmitter to a receiver transmitting the plurality of packets in the priority hierarchy. The real-time data is divided into a plurality of frames. Each of the plurality of frame is divided into multiple packets. Each of the plurality of frames is stored in a frame buffer included within the receiver. During transmission, a level of a plurality of frames included within the frame buffer is monitored. In response to the level crossing a first buffer threshold, a determination is made as to whether a first plurality of the multiple packets are associated with a first priority level.

Abstract: Data is transferred over a bus from one device to another, or between one device and another system resource, such as a central processor. This data is classified into one of several types. "Hard real time" data must be transferred within a specified time limit or "deadline" and it is unacceptable to miss a deadline. "Soft real time" data should be transferred before a deadline and, although some missed deadlines are tolerable, the lower the number of missed deadline the better. "Loss sensitive" data has no deadlines, but any loss of data is unacceptable. "Non-real time" data also has no deadlines, but the lower the time delay in transferring the data the better. The intelligence that controls the transfer of data and schedules access to the bus is distributed throughout the system. Part of this scheduling intelligence is included in the bus arbiter, while the remainder is incorporated in the devices themselves.