Data Center Monitoring and Management Best Practices: How You Can Benefit from Artificial Intelligence and Machine Learning

Editor's Notes

• #4: Speak to what they will walk away from the presentation understanding
• #5: Data Centers efficiency is tied to the original design, based on the expected IT load that operate within them - IT tries to provide facilities and facilities operations with a clear understanding of what rack power density require, today and in the future - The average data center operates for over 20 years. - Operating a data center efficiently over time, is a challenge, and few do it well The challenge with that is: - Accuracy of initial projection - How that changes over time (capacity used, change in IT loads) - Matching the critical infrastructure to the IT load, over the lifespan of operation of the data center requires granular measurement and analytics - Typically customers don’t do this and as a result inefficiencies exist, risks persist, energy is wasted and costs are high A data center operators goals should be to: 1. Deliver on the SLA (availability, and performance) 100% of the time 2. Do so in the most efficient manner that allows for the lowest operational costs possible, with the lowest energy consumption possible – Without risking on delivering on the SLA So What? - Does this matter? Today’s data centers face a challenge that, initially, looks like it’s almost impossible to resolve. Operations have never been busier - analyst projections suggest that workload levels are only going to increase over the next five years – with some projecting a 21% CAGR growth rate between now and 2025 However, critical facilities such as data centers are coming under increased pressure to reduce their energy consumption - particularly as the reality of corporate net zero commitments start to bite These commitments will come under increased focus this week with the UN Climate Change Conference – COP26 – meeting in Glasgow, UK – with the goal of bringing countries together to accelerate action towards the goals of the Paris Agreement and the UN Framework Convention on Climate Change And with corporate data centers already established as one of the world’s highest collective consumers of energy, it’s imperative that IT operations teams do everything they can to deliver the quick carbon reduction wins that will help organizations to deliver on their net zero commitments
• #6: Data Centers efficiency is tied to the original design, based on the expected IT load that operate within them - IT tries to provide facilities and facilities operations with a clear understanding of what rack power density require, today and in the future - The average data center operates for over 20 years. - Operating a data center efficiently over time, is a challenge, and few do it well The challenge with that is: - Accuracy of initial projection - How that changes over time (capacity used, change in IT loads) - Matching the critical infrastructure to the IT load, over the lifespan of operation of the data center requires granular measurement and analytics - Typically customers don’t do this and as a result inefficiencies exist, risks persist, energy is wasted and costs are high A data center operators goals should be to: 1. Deliver on the SLA (availability, and performance) 100% of the time 2. Do so in the most efficient manner that allows for the lowest operational costs possible, with the lowest energy consumption possible – Without risking on delivering on the SLA So What? - Does this matter? Today’s data centers face a challenge that, initially, looks like it’s almost impossible to resolve. Operations have never been busier - analyst projections suggest that workload levels are only going to increase over the next five years – with some projecting a 21% CAGR growth rate between now and 2025 However, critical facilities such as data centers are coming under increased pressure to reduce their energy consumption - particularly as the reality of corporate net zero commitments start to bite These commitments will come under increased focus this week with the UN Climate Change Conference – COP26 – meeting in Glasgow, UK – with the goal of bringing countries together to accelerate action towards the goals of the Paris Agreement and the UN Framework Convention on Climate Change And with corporate data centers already established as one of the world’s highest collective consumers of energy, it’s imperative that IT operations teams do everything they can to deliver the quick carbon reduction wins that will help organizations to deliver on their net zero commitments
• #7: Latest advancements in software, sensor, comms and cloud infrastructures now support real-time optimization at a compelling price point Now imagine the ability to use real-time sensor data to create a digital twin of your data center layout where it can visually represent your current thermal conditions …AND provide recommendations for optimization ….AND know when the changes were made….AND update the digital twin to reflect the changes made…this is the direction where airflow management is headed.
• #10: Very few data center operators have real time visibility to the temperature and humidity of the racks They measure the temperature of the cooling units and occasionally place a few sensors throughout the data center They Hope that the temperatures of the individual racks are all operating within their acceptable range (be that ASHRAE or other) Hope is not a strategy, and every racks thermal performance is unique thermal properties are not transitive between racks, even those that sit side by side Does that matter? Excessive heat impacts server performance and the useful life of the equipment. Server fans run faster, consuming energy and reducing the serviceable life of the server. Servers running hot will begin turning off services. This results in sub-optimal server performance Over-cooled servers result in increased humidity. Humidity is the enemy of the components that the server is comprised of, resulting in shorter useful life of that server The cost of the hardware contained within a server rack can range from $40K to $1M
• #11: The application of AI and machine learning has been applied to optimizing compute, storage, virtualization, and networking, at the software layer for years, and the efficiencies are realized every day. When it comes to operating the data centers themselves, from the critical infrastructure (space power and cooling) to the IT load, (the physical devices that go in the racks), AI and Machine Learning has not yet reached the early adopter stage The emphasis continues to be on designing to anticipated requirements from IT (number of servers, rack power density etc.) The manufacturers of the critical infrastructure continue to do a great job of developing more efficient solutions The server manufacturers continue to release more and more efficient designs, allowing for more compute, more efficiently and more reliably The plan on paper always looks good. But then Reality Strikes……. IT miscalculated the number of servers required IT miscalculated the power density per rack that they would require The roll out schedule of physical servers is much different than anticipated No one foresaw that acquisition, that led to completely unanticipated infrastructure being moved into the data center NOBODY successfully anticipated how technology changes and trends would impact the data center over time
• #12: The application of AI and machine learning has been applied to optimizing compute, storage, virtualization, and networking, at the software layer for years, and the efficiencies are realized every day. When it comes to operating the data centers themselves, from the critical infrastructure (space power and cooling) to the IT load, (the physical devices that go in the racks), AI and Machine Learning has not yet reached the early adopter stage The emphasis continues to be on designing to anticipated requirements from IT (number of servers, rack power density etc.) The manufacturers of the critical infrastructure continue to do a great job of developing more efficient solutions The server manufacturers continue to release more and more efficient designs, allowing for more compute, more efficiently and more reliably The plan on paper always looks good. But then Reality Strikes……. IT miscalculated the number of servers required IT miscalculated the power density per rack that they would require The roll out schedule of physical servers is much different than anticipated No one foresaw that acquisition, that led to completely unanticipated infrastructure being moved into the data center NOBODY successfully anticipated how technology changes and trends would impact the data center over time
• #13: The good news is that the data is there The First Step is to begin Capturing and Measuring the Data This is akin to taking off the blinders
• #14: And when you do…….. The first step is always to admit that you have a problem!!!!!
• #16: Simultaneous monitoring and measurement of the critical infrastructure (power and cooling) and the IT load produces enormous amounts of data Operators have neither the time nor the expertise to interpret calculate and act on that data in a timely manner on a sustained basis No worries, AI and Machine Learning to the rescue….. And the weather looks great!!! Machine learning and AI analytics overcome this challenge. Powerful correlation engines learn, HOW the data center is operating, WHY it is operating that way and WHAT can be done to improve it The actionable recommendations enable the data center operator to make more informed decisions, while maintaining operator control and oversight The AI and M/L algorithms, continue to monitor and continue to observer. This enables ongoing optimization. The system is perpetually looking at ways to improve the environment, while also analyzing the impacts of changes within the environment Continuous Optimization is critical to ensuring that the environment continues to meet the business’s SLA, and in the most efficient manner possible
• #17: Five core requirements for effective AI-enabled data center optimization: Gather accurate cooling, power, and space data at a highly granular level – this means that you can access how your data center is performing in real-time Make it easy to visualize complex data easily and quickly – with comprehensive 3D digital twin visualizations that are easy for data center teams to interpret, enabling the comparison of large data ranges to show changes and highlight anomalies Apply machine learning and AI analytics to provide actionable insights – augmenting measured datasets with machine learning algorithms to provide data center teams with easy-to-understand insights to support real-time optimization decisions Ensure delivery with actionable recommendations for human auditability – providing operations teams with recommended actions for incremental changes that can be easily validated and that will continue to deliver until optimization objectives are met Deploy an ongoing continuous optimization approach – giving data center staff the capability for continuous optimization, supporting them in keeping pace with their ever-changing critical facilities. This can either be through human auditability or setup via automated control systems The result is an AI/ML-powered software-driven optimization approach that helps to change the game for data center operators
• #18: Granular data – every RACK! every CRAC Better 1 sensor per rack than top/middle/bottom on some Wireless & IoT sensors making this affordable Save historical data – learn from occurrences Cloud platforms are a fast & effective
• #20: Example: True cooling capacity – what do you cooling systems actually deliver, is it reducing overtime, possible failure - remove guesswork Do I have an overprovision of IT load Vs Cooling in areas of the room
• #23: DCIM and Traditional Monitoring Solutions do a nice job of tracking the assets within a data center, and can be instrumented to trend data and alert when breaches occur
• #24: Just quickly summarize each of the boxes.
• #28: In 2016, Google and DeepMind jointly developed an AI-powered recommendation system to improve the energy efficiency of Google’s already highly-optimized data centers Their thinking was simple: Even minor improvements would provide significant energy savings and reduce CO2 emissions to help combat climate change. Now they’re taking this system to the next level: instead of human-implemented recommendations, their AI system is directly controlling data center cooling, while remaining under the expert supervision of data center operators. This first-of-its-kind cloud-based control system is now safely delivering energy savings in multiple Google data centers. Reducing energy usage has been a major focus for us over the past  10 years: we have built our own super-efficient servers at Google, invented more efficient ways to cool our data centers and invested heavily in green energy sources, with the goal of being powered 100 percent by renewable energy. Compared to five years ago, we now get around 3.5 times the computing power out of the same amount of energy, and we continue to make many improvements each year. The machine learning system was able to consistently achieve a 40 percent reduction in the amount of energy used for cooling, which equates to a 15 percent reduction in overall PUE overhead after accounting for electrical losses and other non-cooling inefficiencies. It also produced the lowest PUE the site had ever seen.
• #29: They enable the operator to lower risk, and operating costs simultaneously – 100% of the time They reduce energy consumption and carbon emissions – 100% of the time They result in releasing stranded capacity – 100% of the time They achieve this solely by improved operations OF EXISTING INFRASTRUCTURE They reduce the workload on the data center operators They empower the data center operators to make more informed decisions, backed by data The results are easy to see, and easy to measure – Real Time The ROI is measurable, and the time to value can be achieved within a few weeks of initial implementation