Sequential Power-Up Sequence for Computing Cluster PDUs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
High performance computing clusters face issues with inrush currents during power-up, which can trip circuit breakers, and existing methods like adding inductors to power supplies are not fully effective in managing this surge.
Innovation Solution
A method that identifies power distribution units and devices in a cluster, powers them on in a sequence based on ascending commonality of model identification codes to prevent inrush currents from tripping circuit breakers, using a central management entity to communicate with service processors and manage the power-on sequence.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If all devices in a computing cluster are powered on simultaneously, then the cluster becomes operational quickly, but the inrush current can trip circuit breakers at the main distribution center or individual PDUs
Solution Approach 1:
The patent segments the power-on process by dividing devices into multiple groups based on their model identification codes and commonality. Instead of powering on all devices simultaneously, the system powers on devices in sequential batches, with each batch representing a segment of the total device population. This segmentation reduces the peak inrush current at any given moment while still achieving full cluster operational status.
Solution Approach 2:
The patent performs preliminary actions by first identifying all devices in the cluster, obtaining their vital product data including model identification codes, and determining the commonality of each device type before initiating the power-on sequence. This preliminary analysis allows the system to plan the power-on sequence in advance, ensuring that devices are powered on in an order that prevents circuit breaker tripping while maintaining overall productivity.
2Reliability
If inductors are added to the input of each power supply to limit inrush current, then circuit breaker tripping is reduced, but the system complexity and cost increase
Solution Approach 1:
The patent introduces a central management entity as an intermediary between the power distribution units and the devices. This entity coordinates the power-on sequence by receiving information about device commonality, determining the optimal power-on order, and controlling when each device should be powered on. This software-based intermediary approach achieves circuit breaker protection without requiring hardware modifications like inductors, thereby reducing system complexity.
Solution Approach 2:
The patent replaces the mechanical/electrical solution of adding inductors to power supplies with a control-based software system. Instead of modifying the physical power supply circuits to limit inrush current, the system uses a management entity to orchestrate the timing of device power-on events. This substitution eliminates the need for additional passive components while achieving the same reliability goal.
3Reliability
If devices are powered on in sequential batches based on model commonality, then inrush current is controlled, but the time to fully operationalize the cluster increases
Solution Approach 1:
The patent changes the parameter of power-on timing from simultaneous to sequential based on device commonality. By grouping devices with identical or similar model identification codes together and powering on each group in succession, the system controls the inrush current profile. The batch size and timing can be adjusted as parameters to balance between preventing circuit breaker tripping and minimizing total activation time.
Solution Approach 2:
The patent applies partial action by powering on devices in multiple batches rather than all at once or one at a time. Each batch represents a partial completion of the total power-on task, and the process is repeated until all devices are operational. This approach finds a middle ground between simultaneous power-on (excessive action causing inrush current) and individual sequential power-on (inefficient), optimizing both reliability and time efficiency.
Data Source
AI summary
A computer program product includes computer usable program code for: identifying a plurality of power distribution units (PDUs) disposed in a rack, wherein each PDU receives power from a main power source and includes a circuit breaker; identifying a plurality of devices disposed in the rack, wherein each device receives power from one of the PDUs, and wherein the plurality of devices are selected from server nodes, network switches and external data storage devices; obtaining vital product data from a service processor in each device, wherein the vital product data identifies the device by a model identification code; and powering on, for each of the PDUs, the plurality of devices that are connected to the PDU in a sequence to prevent an inrush current from tripping the circuit breaker within the PDU, wherein the sequence powers on devices in order of ascending commonality of the model identification code.


