Another way cool approach to Data Center build and operation presented their approach to our group at the same time. Open19 has created a Lego type approach to the data center, dropping the delivery time to build a rack of components from roughly 8 hours per rack down to 2.
So, how did they do this? If you imagine that rack components all fit into the 19” standard platform, then you can break down the properties of each individual type of component into any of 4 standard sizes. There are half-width, 1 RU servers, half-width 2RU servers, Full width 1RU, and Full Width 2U devices. What we’ve got are shelved devices, which all slide into pre-sized slots in the rack, and these slots are part of a whole lego type approach. So, you give your rack a couple different flavored slot shelving units, which most appropriate align with the profile of the servers, switches, power, etc. you require, and then fill them with the components you need.
With rack efficiency at 96%, not even the Open Compute Project or Olympus can compete. Cooling is so efficient that no additional fans are required, to take up more rack space. As much cabling is built into the back plane as possible, such that the spaghetti mess that often confuses the builds, and time taken to do the racking and stacking let alone, the troubleshooting of potential problems is simply taken care of on a plug and play basis.
This architecture is not for everyone, mind you. If your organization has legacy equipment that does not fall into the standard architectures required by Open19, or if the number of racks to implement isn’t at “RackScale” this probably isn’t for you. If, however, you’re building out a data center newly, and need 100,000 devices, deployed at scale, rapidly, efficiently, and with consistency, you’d find yourself in good stead by looking at this approach very closely.
Mind you, the servers and other gear are quite standard today. These are builds that the likes of HPE, SuperMicro and Quanta are building consistently. Therefore, if you desire this kind of scale, you’re really not so limited in your X86 vendor of choice.