Browse Prior Art Database

System for Load Prediction Based on Micro Topic Interactions Disclosure Number: IPCOM000248675D
Publication Date: 2016-Dec-24
Document File: 3 page(s) / 176K

Publishing Venue

The Prior Art Database


Disclosed is a system for load prediction based on micro topic Interactions. The system monitors a combination of users and loads by measuring given parameters and then takes appropriate action to ensure load balance.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 47% of the total text.


System for Load Prediction Based on Micro Topic Interactions

Different costs are associated with using a system, and the system has to balance cost with expected performance. For example, a social media website must decide which data should be pro-actively moved from the capacity disk to the performance disk at a given time. The same general scenario and questions exist for different layers in the network, central processing units (CPUs), and application posture.

The state-of-the-art solutions for throttling resources are based on measurements and so employ a reactive approach. However, machine learning methods systems can predict future load using user and application information from a social network. For example, a system can detect events on a social network that have a human crowd magnitude that is influencing power or network requirements in an area.

Unfortunately, that approach is not granular enough. With tight profit margins, consider an on-premise or cloud provider for instant messaging: how can that person infer load/resource requirements from minute-to-minute in order to meet cost constraints?

Load estimation for a system should take into account the differential load created by the combinations of active users. Currently, if user1, user2, and user3 are active, then the load on the system is likely to be a simple calculation of the associated likely load. However, if user2 is not active for the discussion of a specific topic, then that may have an increased impact on the load. For example, when user2 is not available, the average load on the system from user1 and user3 dramatically decreases, so the prediction algorithm considers those. Those combinations can be analyzed across the network by topic.

Based on the above, a system can take various actions, which are not novel, such as better pre-caching, management of memory, adding/removing extra virtual machines, (VMs), etc.

The novel idea is for a system to monitor a combination of users and load by measuring the following:

· Topic · Group of users · Node-to-node interactions · Who is online and expected online availability · Calendaring and scheduling · Historical load used based on the above

For a given topic (e.g., “j2ee”, in the figure below) in a calendaring and scheduling event, the system uses historical analysis to understand the likely reach/impact on the system by modelling against the likely availability of the involved parties. As shown in the figure, the system can ascertain the impact.

Figure 2: Example for a calendaring/scheduling event


The above table also holds the time lag information between each of the node-to-node interactions (e.g., there is an average delay of two minutes between "A->C" and "C->E" interactions).

Therefore, at 21:17 on Thursday, the system ascertains that if the topic “j2ee” is initiated, then the node-to-node (or ripple effect) will be initiated.

The system ascertains a topic, extracting and providing the start and end times for each