5 results about "Offered load" patented technology

The present application generally relates to systems and methods for detecting network failures for machine learning workloads. Aspects of the present disclosure relate to detecting a slow fault condition in a network executing a machine learning application. The detection may be performed by comparing the load provided by the machine learning application to network utilization over a short time interval. The load provided by the machine learning application (sometimes herein referred to as provided load) may be measured by provisioning the machine learning application to share data upon request. For example, the application may share data, such as an amount of information to be sent associated with the request, at a start time at which the request starts and an end time at which the request ends.

The utility model discloses a load test tool and a load test system. The load test tool provided by the utility model comprises an input end, an output end, a power resistance network and a current detection device, the power resistor network is located between the input end and the output end and comprises a first main circuit, the first main circuit comprises a first load unit, and the first load unit comprises a first switch, a first high-power resistor and a second high-power resistor; the first switch is used for switching one of the first high-power resistor and the second high-power resistor to conduct the input end and the output end; and the current detection device is connected in series with the power resistor network and is used for detecting a current value between the input end and the output end. Therefore, only one of the first high-power resistor and the second high-power resistor needs to be switched through the first switch to conduct the input end and the output end, the resistance value of the analog load can be adjusted, the testing efficiency is improved, and meanwhile the current detection device improves the monitoring timeliness and sensitivity of the current value.

The application discloses a kind of distributed optimal power flow end-to-end solving method and device, method includes: distribution network is divided into several sub-regions, and establishes distributed optimal power flow problem;Real-time load data is input into trained ADMM-Net;The ADMM-Net includes input layer, variable estimation layer, differentiable convex optimization layer and output layer, input layer is used to receive load data;Variable estimation layer extracts system potential operating characteristics based on the load data provided by input layer, and outputs the preliminary estimate of global variable and Lagrange multiplier;Differentiable convex optimization layer is based on the preliminary estimate of global variable and Lagrange multiplier output by variable estimation layer to execute once ADMM iteration, generate updated global variable and Lagrange multiplier value;Output layer is based on the iteration result of differentiable convex optimization layer, and outputs distributed optimal power flow solution.The application can greatly reduce the time required for solving distributed optimal power flow.

The system and method, device and electronic equipment provided by the present disclosure comprise a forwarding plane, a session plane and a monitoring plane, the monitoring plane is used for monitoring state information of the forwarding plane and the session plane; the forwarding plane is used for obtaining to-be-forwarded traffic and transmitting forwarding information carried in the to-be-forwarded traffic to the session plane; the session plane is used for responding to the forwarding information, matching configuration information corresponding to the to-be-forwarded traffic from a preset database according to the forwarding information, and performing interaction of session and configuration with the forwarding plane, and transmitting the configuration information to the forwarding plane; the forwarding plane is further used for receiving the configuration information and performing forwarding processing on the to-be-forwarded traffic according to the configuration information and the state information. Compared with the related art, the embodiment of the present disclosure can avoid forwarding the to-be-forwarded traffic to the failed node and reduce the probability of service interruption by modularizing and splitting, monitoring the state information of each node of the system through the monitoring plane module, so that the system has continuity and high availability.

A data processing system and method for reallocating inroute channels of a receiver of a data communication system, the receiver being configured to receive communication signals of the data communication system on the inroute channels. The reallocating is performed by detecting an offered load of the communications signals on the inroute channels of the receiver, reallocating Scrambled Code Multiple Access / Asynchronous Scrambled Code Multiple Access (SCMA / ASCMA) channels of the inroute channels to Time Division Multiple Access (TDMA) channels of the inroute channels upon detecting that the offered load is less than a predetermined threshold, and reallocating TDMA channels of the inroute channels to SCMA / ASCMA channels of the inroute channels upon detecting that the offered load is greater than the predetermined threshold.