It has been almost four days since I last blogged. Have been busy with a client in Tokyo.
This photo is a view from my hotel room at Shibuya. On the top left, you can see Tokyo tower at a distant away.
Look carefully at the picture and you’ll see that in general, there are buildings which are 10 stories high, in the middle of smaller buildings and there are occasionally a few which are 20 or more stories high. And here I am, looking out of my hotel room window, on the 29th storey. This is exactly the same state that we most likely will observe in a typical data center floor. There will be racks with highly densed equipment, consuming upwards of 15kW or more power, and right next to it, might be a patch panel rack with a few layer 2 switches, which is lower density. Each row of equipment racks and each individual rack will have varying composition of density.
Therefore, from an operational angle and from an efficiency angle, it is no longer effective to apply a “one-size-fits-all” design for data center facilities of the future. It is necessary to adopt a very granular approach in designing features to address the variance. These could be in the form of:
- Localizing cooling, i.e. bringing the cooling closer to the individual racks, hence achieving the ability to design appropriately sized cooling capacity to address high density racks and low density racks, regardless of their location. Traditional ambient / room cooling architecture would have too much limitation to address the variances.
- Flexible power distribution instead of hard fixing power distribution to each rack location. Refer to the idea of power distribution busways I shared earlier. Pre-laying power cables and whips across the facility is going to fix the location of your equipment racks and when the load are higher than anticipated, it creates a need to change and add additional capacity to that location. Such changes will be a hassle and bring about longer lead time in provisioning new services.
- Scalability – If you consider your plumbing system in the data center, whether it is for refrigerant or chilled water, there are a few points to think about. Firstly, installation of these plumbings generally would require much effort and most of the time, hot works too. Secondly, pipe sizes are finite and once installed, you can’t change these sizes. Therefore, any additional loads in the future are constrained with physical limitations. It is always better to think about potential future loads and upsize it now.
- Moderation and modularity in key infrastructure. Often times, it is not true that “the bigger, the better”. Limit losses from your key infrastructure components by deploying modular units gradually and when your capacity grows beyond a preset threshold.
That’s all my thoughts, greetings and salute from Tokyo. Have to get back to work now…
Tags: Architecture, Data Center, Density, Efficiency, PUE