Abstract: Some embodiments provide a method for a load balancer executing in a public cloud to load balance requests to datacenters from a network management system implemented in the public cloud. The method receives a message from a network management service for a local network manager at a particular datacenter of multiple datacenters managed by the network management system. The method maps a datacenter identifier for the particular datacenter included in the request to a particular connection server that handles a persistent connection between the local network manager at the particular datacenter and the network management system. Persistent connections between the datacenters and the network management system are handled by multiple different connection servers. The method sends the message to the particular connection server for the particular connection server to transmit the message to the local network manager via the persistent connection.

Abstract: Some embodiments provide a method for initiating a connection between a network management system in a public cloud and a datacenter managed by the network management system. The method is performed at a controller executing in the public cloud. The method receives a registration message from a particular datacenter to be managed by the network management system. The method identifies one of multiple connection servers in the public cloud for handling a persistent connection initiated by the datacenter. The method configures a load balancer in the public cloud to statically select the identified connection server for data messages that include a datacenter identifier associated with the particular datacenter.

Abstract: Some embodiments provide a method for a connection server that handles connections between a network management system in a public cloud and datacenters managed by the network management system. The method receives (i) a first application programming interface (API) request for a first local network manager located at a first datacenter belonging to a first datacenter group of a first tenant and (ii) a second API request for a second local network manager located at a second datacenter belonging to a second datacenter group of a second tenant. Based on site identifiers respectively associated with the first and second datacenters and included in the first and second API requests, the method respectively sends the first and second API requests to the first and second local network managers via first and second connections previously initiated by the first and second local network managers.

Abstract: Some embodiments provide a method for a connection server that handles a connection between a network management system in a public cloud and a datacenter belonging to a datacenter group managed by the network management system. The method receives an API request, having a first format, for a local network manager located at the datacenter. The method translates the API request into a second format used by the connection between the network management system and the datacenter. The connection is a type that (i) allows for the connection server to push data onto the connection when the connection is initiated by a connection agent at the datacenter and (ii) uses the second format. The method sends the translated API request in the second format to the connection agent via the connection. The connection agent translates the API request into the first format for execution at the local network manager.

Abstract: Some embodiments provide a method that identifies a first number of requests received at a first application. Based on the first number of requests received at the first application, the method determines that a second application that processes requests after processing by the first application requires additional resources to handle a second number of requests that will be received at the second application. The method increases the amount of resources available to the second application prior to the second application receiving the second number of requests.

Abstract: Dynamically scaling control plane for ingress services for large numbers of applications with minimal traffic disruption includes receiving an estimate of a number of applications to be executed by multiple clusters implemented by an orchestrator platform. Each cluster includes multiple containers. The multiple clusters implement a centralized controller that control execution of the applications by the multiple clusters. The centralized controller is sharded into a variable number of controllers that collectively control the estimated number of applications based on the estimate of the number of applications and a pre-determined number of applications that each controller can control. Each controller of the variable number of controllers controls an execution of a respective subset of the applications. In response to a change in the number of applications over time, the number of controllers is modified based on a number of applications to be executed by the multiple clusters at any given time.

Abstract: Some embodiments provide a method that identifies a first number of requests received at a first application. Based on the first number of requests received at the first application, the method determines that a second application that processes requests after processing by the first application requires additional resources to handle a second number of requests that will be received at the second application. The method increases the amount of resources available to the second application prior to the second application receiving the second number of requests.

Abstract: Some embodiments provide a method for scaling a service chain that includes multiple services, each of which is provided by one or more instances of the service. The method identifies that a first service in the service chain has received a number of requests. For each service in the service chain, the method (i) identifies a scaling factor that estimates a portion of requests received at the first service that will be subsequently received at the service and (ii) deploys a number of additional instances of the service based on the identified scaling factor for the service and the number of requests received at the first service.

Abstract: Dynamically scaling control plane for ingress services for large numbers of applications with minimal traffic disruption includes receiving an estimate of a number of applications to be executed by multiple clusters implemented by an orchestrator platform. Each cluster includes multiple containers. The multiple clusters implement a centralized controller that control execution of the applications by the multiple clusters. The centralized controller is sharded into a variable number of controllers that collectively control the estimated number of applications based on the estimate of the number of applications and a pre-determined number of applications that each controller can control. Each controller of the variable number of controllers controls an execution of a respective subset of the applications. In response to a change in the number of applications over time, the number of controllers is modified based on a number of applications to be executed by the multiple clusters at any given time.

Abstract: The present invention discloses a system for selection of candidate video frames from traffic video comprising an imaging processor operatively connected to traffic video source. The imaging processor receives the traffic video from said source for selection of candidate video frames from the received traffic video for analysis of entire traffic video stream to extract vehicles details from therefrom including detection of situations of moving vehicles and non-moving vehicles. The traffic video is forwarded to the imaging processor from the source at an appropriate rate suitable for said candidate frames selection computation.

Abstract: The present invention discloses a system for selection of candidate video frames from traffic video comprising an imaging processor operatively connected to traffic video source. The imaging processor receives the traffic video from said source for selection of candidate video frames from the received traffic video for analysis of entire traffic video stream to extract vehicles details from therefrom including detection of situations of moving vehicles and non-moving vehicles. The traffic video is forwarded to the imaging processor from the source at an appropriate rate suitable for said candidate frames selection computation.

Abstract: Systems and methods are provided for changing the number of replicas of a pod based on the size of the message queue. A method may include determining a number of replicas of a pod; determining an owner threshold for a size of a message queue of the replicas of the pod, the size representing a number of messages waiting in the message queue to be processed by the replicas of the pod, the owner threshold being provided by an owner of the pod; occasionally determining, for the replicas of the pod, the size of the message queue, and a rate of change of the size of the message queue; and responsive to determining the size of the message queue, changing the number of replicas of the pod based on the determined size of the message queue, the determined rate of change, and the owner threshold.