System Efficiency Optimization
Application and system efficiency optimization is the goal of intelligent monitoring and control of ASGC operation. Started from the power and thermal efficiency problems of ASGC data center, detection of refrigerant operating issue and abnormal components based on machine learning technologies have been in production. Anomaly detection on compute, storage, network and security systems are the next targets.
Before all of that, design and prototypes of fanless single rack data center as well as smart power control of IT facility had been implemented.
4. Computing System Efficiency
5. Storage System Efficiency
6. Networking System Efficiency
Fanless Single Rack Data Center
Replacement of the current air convection-base cooling in Data Center and IT facility by conduction-based approach could reduce at least 70% power consumption in cooling system and 15% power consumption of IT equipment itself as well as the acoustic noise level of over 90dB. A novel conductional colling architecture based on space technology has been developed and is patented in US, China and Taiwan.
Accordingly, a full-rack prototype with servers, storages and network switches achieved ~1.2 PUE in summer time.
The outcome is an noise-free, high PUE and easy installation single rack data center which could be served perfectly as the building block of various size data center.
A new plan to achieve PUE<1.1 is also composed to further optimize the power and termal efficiency of a data center.
System Efficiency Optimization – Smart Power Control for IT Facility
IT equipment in a cloud center is not always under a full load in the real condition. A standby server could still take up to 50% of its maximum power consumption.
ASGC has developed an automatic and smart power control mechanism for IT equipments, which includes the prediction of computing resource requirement, power and IO monitoring, virtual machine/container deployment and error recovery. The goal is to turn into appropriate power states (e.g., freeze/suspend/hibernate modes) according to the upcoming job characteristics flexibly and intelligently so that power can be saved as much as possible with mimimal performance degrade.