Energy Proportional Multiprocessor Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Datacenter networks consume a significant amount of power regardless of utilization levels due to 'always on' high-speed channels, leading to inefficiencies and high operational expenditures, as they cannot be quickly deactivated or reactivated without affecting data flow and routing algorithms.
Innovation Solution
Implementing independent flow control on unidirectional links in a multiprocessor system to dynamically tune link speeds based on traffic intensity, using a flattened butterfly topology and credit-based link-level flow control to estimate bandwidth needs and adjust link speeds in real-time, matching power consumption with traffic demand.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If high-speed channels operate in 'always on' mode, then data transmission performance is maintained, but power consumption remains high regardless of utilization level
Solution Approach 1:
The patent implements dynamic link reconfiguration that allows channels to transition between active and inactive states based on traffic conditions. The system periodically estimates bandwidth needs and dynamically tunes link speeds or deactivates links when idle, transforming the static 'always on' architecture into a dynamic system that adapts to varying workload demands.
Solution Approach 2:
The system employs periodic bandwidth estimation and link reconfiguration cycles. Rather than continuously monitoring or maintaining constant state, the system performs periodic assessments of traffic patterns and adjusts link states accordingly, balancing the need for performance availability with power savings during low-utilization periods.
2Use of energy by moving object
If links are deactivated to save power, then energy consumption is reduced, but routing algorithms perceive links as faulty requiring workarounds
Solution Approach 1:
The patent segments the link state management by treating unidirectional channels independently. Each unidirectional link can be deactivated without affecting the other direction, allowing granular control over power consumption while maintaining routing simplicity. This segmentation avoids the need for complex bidirectional coordination.
Solution Approach 2:
The system introduces an intermediary mechanism where the routing algorithm is informed of link deactivation status through credit-based flow control signals. This intermediary information channel allows routing decisions to be made without direct coordination between all nodes, simplifying the complexity while enabling power savings.
3Use of energy by moving object
If link speeds are dynamically tuned, then power consumption matches traffic demand, but reactivation latency and negotiation overhead increase
Solution Approach 1:
The system performs preliminary actions by maintaining link configuration information and negotiation state in memory during inactive periods. When reactivation is needed, the system can resume operation more quickly because the negotiation parameters and alignment information are preserved rather than requiring complete re-establishment from scratch.
Solution Approach 2:
The patent implements parameter changes in link speed and operational state based on estimated traffic intensity. The system adjusts data rates, symbol alignment, and lane configuration parameters dynamically, allowing the system to optimize power consumption by operating at lower parameters during low-utilization periods while maintaining the ability to quickly transition to higher performance modes when needed.
Data Source
AI summary
Energy proportional solutions are provided for computer networks such as datacenters. Congestion sensing heuristics are used to adaptively route traffic across links. Traffic intensity is sensed and links are dynamically activated as they are needed. As the offered load is decreased, the lower channel utilization is sensed and the link speed is reduced to save power. Flattened butterfly topologies can be used in a further power saving approach. Switch mechanisms are exploit the topology's capabilities by reconfiguring link speeds on-the-fly to match bandwidth and power with the traffic demand. For instance, the system may estimate the future bandwidth needs of each link and reconfigure its data rate to meet those requirements while consuming less power. In one configuration, a mechanism is provided where the switch tracks the utilization of each of its links over an epoch, and then makes an adjustment at the end of the epoch.