Carbon Relay wants to fire up the data center industry by using artificial intelligence to help IT pros cool down these big facilities.
The Boston newcomer released two products this week that use AI to automate cooling systems and integrate those systems with core IT assets, such as server and storage hardware and software, to improve data center energy efficiency.
The two applications, Optimize and Predict, work with an IT organizations' existing network of sensors, IoT platforms, large data sets and data visualization tools to automate the control of industrial heating, cooling and air conditioning (HVAC) systems to boost their efficiency. The software can also help predict when the systems need maintenance, and also offer failure detection, outages and know when to redistribute computing resources.
In addition, Carbon Relay executives said they believe these efficiencies through automation will lead to less human intervention, which could also drive down costs in large corporate data centers.
"Typically, you would need IT personnel walking around turning the dials to adjust the temperatures," said Matt Provo, co-founder and CEO of Carbon Relay. "Our [software] agent is making all those adjustments and doing so continuously."
The software makes hundreds of quick decisions to support predictive capabilities across many variables. For instance, if the Carbon Relay system knows the outside temperature will rise 13 degrees, it can turn the temperature down 3 degrees in a data center "hot aisle" or turn it up 2 degrees in a cold aisle.
The Optimize software can simulate a customer's data center in 48 hours, based on the information it collects using only a facility blueprint and electrical layout diagrams, Provo said. The product then tunes up the simulated environment to move into production in about 12 weeks. It does so without any connection to IT, IoT or sensors, he said.
The Predict software analyzes a user's historical operations data, including temperatures and power levels, which it does collect from IoT sensors inside a data center. The product uses the data to train its predictive model.
Data visualization tools help data center managers discover and prioritize areas that can be more efficient.
Our APIs are designed to send out floats or values, which trigger actions to be taken across 37 different variables.
Matt ProvoCEO, Carbon Relay
Carbon Relay tailors its software to be compatible with major IoT platforms, including Siemens MindSphere and Cisco Kinetic along with data center management as a service (DMaaS) products such as Schneider Electric, and data center infrastructure management (DCIM) software such as Maya Heat Transfer Technologies. This is critical to move the company's intelligent agent into production.
"Our APIs are designed to send out floats or values, which trigger actions to be taken across 37 different variables," Provo said. "These variables can be environmental, or things on the IT side of the house, or be related to HVAC equipment," he said.
The Carbon Relay technology exposes an actuator, a small piece of hardware in nearly every piece of industrial machinery that serves as a control point, to integrate with an IT shop's existing network of sensors, IoT platforms and software. Once exposed, the company's agent software can control how the actuator operates with a range of different computer and HVAC systems. Moreover, if users have already bought industrial automation software or an IoT platform, the agent can mesh with those too.
This ability to work with a shop's existing IT equipment can help lower costs by sidestepping switching costs with items like sensors and IoT platforms, Provo said.
Carbon Relay's systems are available as both on-premises software and as a service. The former requires direct connection to the machinery's set points/actuators or to expose them via an IoT platform integration, according to the company. Carbon Relay wouldn't disclose pricing, but said it is calculated based on a facility's square footage, HVAC tonnage capacity and electrical consumption and utilization.
Screenshot of Carbon Relay Optimize software
Fill the gaps in DCIM, DMaaS
Carbon Relay offers a practical, more user-friendly approach to what is often a complex implementation, said Rhonda Ascierto, vice president of research at the Uptime Institute. The ability to quickly stand up a simulated data center environment could hasten adoption.
"I like the idea it can fully simulate in 24 hours a nonproduction version of a data center," she said. "I think it helps users to more easily get their arms around this."
The company is also smart to supplement DCIM platforms rather than replace them. "They aren't trying to be DCIM or DMaaS, although they need those platforms to get into production environments," she said.
Carbon Relay's approach also looks beyond the cooling and power issues that enterprise-level data centers face, and considers related issues with server and storage hardware and software.
"They are taking more of a holistic approach to the data center than some [competitors]," she said. "California Data Science and Solecular are the other 'pure plays' focused on AI in the data center, but I believe they will all have to partner with DCIM and DMaaS companies."
One barrier to adoption of AI in general, which could also apply in the case of Carbon Relay's offerings, is what Ascierto called a "black-box concern" -- that the complexities and increasingly autonomous outcomes from AI become harder to understand.
"This could raise questions about visibilities and accountability ... particularly for third-party data center service providers, including colos and around SLAs [service-level agreements]," she said.
Carbon Relay plans to sell directly to customers and add reseller partnerships later. The company has been in development for the past three years; its primary financial backer has been Taiwan-based Foxconn Technology Group, which is also testing the technology, according to the company. Other IT shops are now in evaluation, the company claims, and it hopes to add DCIM partnerships within the next few months as well.
