Private data centers, including hyper-scale, have the advantage of being able to skip rack or cabinet doors altogether or leave them unlocked for faster access. But in colocation or hosting facilities, where multiple customers and vendors have access, cabinet doors are important for security. Some customers even audit when doors are opened, because often even the mix of equipment in a rack is confidential, and using secure cabinet doors avoids the expense of caging off areas. But those doors add extra challenges to monitoring cooling efficiency in the data center.
“We know virtually everything we need to know, telemetry-wise, about what's going on on the data centre floor, what kind of air we're presenting to the cabinets, and what kind of heat is being produced, and what is emanating; but we don't have a good window into the climate dynamics inside the cabinet itself with the doors closed,” Hank Koch, senior VP of mission critical facilities at OneNeck IT Solutions pointed out in an interview with Data Center Knowledge. (The company runs several multi-tenant data centers, including colos and managed services.)
“If you open the door to figure that out, then you’ve you completely changed the thermal dynamics and the fluid dynamics of the air,” he said. “That’s no help getting to the source of any heat radiation issues if you’re investigating interactions between a very hot chassis device and a very cold one, or a more sensitive one nearby.”
Temperature readings from elsewhere in the data center, or even from the CPU, don’t help you figure out the air temperature at different locations inside a cabinet. Many customers aren’t comfortable having data center operators putting wired sensors inside their cabinets – but they will blame the hoster for any heat issues in those cabinets.
To deal with this problem, the team at OneNeck’s Iowa data center developed a robotic sensor probe that fits onto a standard cabinet door with no drilling or special mounting hardware required, controlled externally by Bluetooth and moving up and down a belt-driven rail to collect temperature data for each rack in the unit and create a full heat map.
“The idea was to come up with a measurement device that’s not obtrusive, that doesn’t obscure any airflow, and that doesn’t invade the cabinet wiring, but can help a customer determine the climate dynamic in the cabinet in case they want to be more proactive about how they arrange things, wire things, or diagnose things related to layout within the cabinet,” Koch said.
That might be as simple as moving units that generate more heat further up the rack, so their waste heat isn’t warming otherwise cooler components above them, or moving wiring that's impeding the airflow. Often, he said, customers just aren’t aware of the impact of heat radiation from a hot unit on neighboring components, or what a difference cable layout makes.
Air circulation inside a cabinet can often be improved. “On the air-intake side of the cabinet, if you’re not religious about [fitting] blanking panels, you can find you’re creating a circulatory system where hot air that’s supposed to be going out the back is ingested at the front by cavities between the ‘U levels’ in the cabinet,” Koch explained. “Some units are deeper, some units are not as deep, and that can create cavities and places where hot air can go that's unintended, and it can find its way back to the front and heat up devices. The data center may be providing wonderful cool air, but inside the cabinet, air is recycling from the hot to cold side and being re-aspirated by a unit that should be getting much cooler air.”
A surprisingly common problem is network and telecoms components being installed the wrong way around – which may be more convenient for cable access but results in fans blowing hot air in the wrong direction. Convincing customers this is a problem can be difficult until they get component failures from overheating, Koch noted, speaking from experience.
Managing temperatures inside cabinets could reduce cooling costs and avoid equipment failures, including flagging problems like power supplies that run too hot.
The heat map will be useful in surveying existing cabinets for issues and making sure new cabinets are set up correctly. (And it’s easy to move from cabinet to cabinet). If workloads are static, it might not be necessary to keep the sensor in a cabinet long-term, although it would also be useful to monitor the impact of maintenance, which often causes failures a couple of weeks later if cabling has been disturbed.
“With colocation, all that side of the cabinet is the customer’s. We can have all the best practices we can on our side, but you have customers working inside the cabinet, sending in their own people or vendors working inside the cabinet,” Koch pointed out. “You have this post-implementation discovery [of problems]; it would be much better to be able to detect that in the first 24 hours.”
The OneNeck robot sensor fits well into the trend of making data center cooling increasingly granular and engineered. “Up until now, most data centers have done their cooling by brute force,” Koch said. “We run data centers based on data and science and engineering, and up until now, as far as I’m concerned, without a device like this you don't have science and engineering and data inside a cabinet. What you have is hunches. You have a few readings here and there, not a full heat map. This provides a heat map with quantitative data to help you get evidence.”
The first version is designed for cabinets, but the underlying technology could be used in a range of locations. “If it gets the right electrical rating, it could be in a transformer, or you could hang it on the wall of a data room to get a sense of what the thermal clines of heat are from floor to ceiling,” he suggested.
While OneNeck customers will be the first to be able to take advantage of the new robot, the company is also in discussion with manufacturers who might make it more widely available, because so many multi-tenant facilities would find it useful.