By Marlow Weston
Sustainability, defined in the Oxford dictionary as “avoidance of the depletion of natural resources in order to maintain an ecological balance,” is an immediate need. Energy is rapidly becoming a precious commodity in both financial and environmental costs. While we have historically considered ways to reduce power consumption in every-day items, such as low-power light bulbs, we are now also searching for ways to reduce costs in datacenters, which have a large draw on local resources.
Depending on your source, datacenters currently consume between one and three percent of the world’s energy. This number continues to grow as we demand more compute in everyday tasks. We want computers to take responsibility for choosing the optimal path between destinations so we can focus on other minutia of our travels. We want computers to enable quick design of new vaccines that are more effective during our current COVID pandemic. We want computers to enable even faster design of vaccines to protect us from the next pandemic. We want to be able to know what the weather is three weeks out for our next outing. We want cars to drive us between places so we can finally get those naps that we remember fondly from our childhoods. These desires require the use of data-driven analytics, which causes an increasing demand for servers.
Over half of the companies surveyed in “Renewable Energy Alone Can’t Address Data Centers’ adverse Environmental Impact” reported using only between 20 and 40% of their compute power. If we could power down the remaining machines when not being used, they could be saving between 80% and 60% of the power consumed.
Realistically, powering down machines entirely is not always practical. There can be unforeseen consequences with continuously adding and removing machines, such as added latency while placing machines online and outright failure of boxes booting. We can, however, optimize the resources we have by reducing power consumption of powered systems. It is a rare to have such a straightforward opportunity for doing well by doing good by embarking on a quest to maximize performance per watt.
We have designed the Kubernetes* Power Manager to expose and utilize power management technologies in a Kubernetes environment. We use Intel® Speed Select Technology (Intel® SST), which is a powerful collection of features that offer granular control over CPU power consumption on a per-core basis. This fills the gap between container orchestration and hardware feature enablement. The Kubernetes Power Manager currently supports Intel® Speed Select Technology – Base Frequency (Intel® SST-BF), Intel® Speed Select Technology – Core Power (Intel® SST-CP), and Frequency tuning.
- Intel SST-BF (Speed Select – Base Frequency) allows users to control the base frequency of particular cores. This guarantees a level of performance on the CPU. Priority cores can be set to a higher base frequency than other cores on the system, so the CPU-intensive workloads can be scheduled on performant cores while other, less demanding workloads can be placed on the lower-frequency workloads.
- Intel SST-CP (Speed Select Technology – Core Power) allows users to group cores into levels of priority. This feature allows the user to group cores in levels of priority. If there is power to spare on a system, the power manager does its best to distribute on cores based on their priority level.
- Frequency Tuning allows users to change frequencies of individual cores on the system. This allows users to hand-tune how much power their workloads are consuming.
Power management is one of the many open source projects Intel supports in our efforts to enable the open software ecosystem. In addition to our efforts in power management, there is a new initiative within the CNCF* forming, called the Environmental Sustainability Working Group, that will help guide the community to define environmental sustainability factors to reduce the carbon footprint in Cloud Native infrastructure. Intel is one of many active participants in this new initiative. Please join us on this exciting new journey.
##
***To learn more about containerized infrastructure and cloud native technologies, consider joining KubeCon + CloudNativeCon Europe 2022, May 16-20.
ABOUT THE AUTHOR
Marlow Weston is currently a Cloud Software Architect at Intel working on resource management. Previously, she has worked in a variety of areas including MLOps, high performance computing tools, security, embedded systems, kernel drivers, tracing libraries, and web scrapers. She began adult life by pursuing a degree in chemical engineering, but in the process realized that programming was more fun than solving piles of differential equations. After completing her degree, Marlow wandered off into the dark universe of computer science, never to return.
References:
[1] Data Centres and Data Transmission Networks. George Kamiya, Iea.org; https://www.iea.org/reports/data-centres-and-data-transmission-networks
[2] Renewable Energy Alone Can’t Address Data Centers’ Adverse Environmental Impact. Asaf Ezra, https://www.forbes.com/sites/forbestechcouncil/2021/05/03/renewable-energy-alone-cant-address-data-centers-adverse-environmental-impact/?sh=6b923b565ddc
[3] Introducing Doing Well by Doing Good, Anand Sampat, medium.com; https://medium.com/dwdg/doing-well-by-doing-good-fact-or-fiction-8ea9a6fcccfd
[4] Kubernetes Power Manager; https://github.com/intel/kubernetes-power-manager
[5] Charter Environmental Sustainability WG; https://docs.google.com/document/d/1JaF7lSUmLQ2zmScmca6UF7PgbjMzSxjhhjx2LThThaY/edit#heading=h.x77xxicolnig
[1] Open Intel; https://www.intel.com/content/www/us/en/developer/topic-technology/open/overview.html
Notices and Disclaimers
Intel technologies may require enabled hardware, software or service activation.
No product or component can be absolutely secure.
Your costs and results may vary.
© Intel Corporation. Intel, the Intel logo, and other Intel marks are trademarks of Intel Corporation or its subsidiaries. *Other names and brands may be claimed as the property of others.
