In our daily life, we deal with recurring capacity problems day in and day out. Planning for capacity allocation, whether in our personal lives or at work, is a mix of science, art, guess works and perhaps some level of luck too. This is the same whether it is for sizing up physical compute and data arrays for a virtualized environment and likewise, for data center facilities.
Michael Manos wrote an interesting piece around data center capacity planning (entitled “Chiller-side Chats: The Capacity Problem“) and provided a very impartial explanation to bridge the understanding between IT, data center manager and the business around this complex subject. In Manos’ blog, he elaborated using an example of planning the power allocation and distribution in a data center.
If we analyze further the challenges in the example that Manos cited, planning for reserved power and the maximum power to be allocated per rack in the data center is made even more complex in the long run. Lets take a look at two possible scenarios:
- As time goes by, IT equipment will age, deteriorate, run out of compute capacity or go past its useful life. It will eventually reached its EOL (End-of-Life) state. When that happens, these equipment gets decommissioned, removed and replaced with new equipment which could be in all sort of form factors. These new equipment may not have the same power rating and consumption as the obsolescent equipment, adding to the risk of load over-drawing beyond the maximum allocated power.
- The other scenario where it is more challenging is an unpredictable IT environment. This applies to colocation and hosting service providers. They’d never know who will be their next customer and would almost be impossible to plan for the next batch of IT equipment that will be coming through their doors and going into their data center.
Well, if we look at the problems usually faced in planning for power capacity, the difficulty is in nailing down actual consumption and allocated power. In a data center where power is distributed to each IT equipment racks through a series of under floor cables, or whips, and fed from a distribution panel, which is fed by a secondary breaker typically from a PDU. So, under-provisioning would mean having to run additional cables under the floor to accommodate the additional load, creating more cables under the raised floor and more headaches for the data center manager.
So, what are the ways we could, at least, reduce the risk of over-provisioning or under-provisioning?
One of the emerging way to address this problem is to use a flexible electrical busway. These busways are basically a track system with a continuous open access that allows power plug-in units to be inserted anywhere along the busway. The design incorporates 3-phase systems consisting of electrified copper conductor bars within a lightweight and safe housing. The plug-in units allows insertion and removal of circuits without any shutdown of the power. In comparison, this modular method allocates power for a group of IT racks, rather than for one single rack.
Therefore, the allocated capacity can be more flexibly managed as consumption can be shared between multiple racks. In Manos’ example earlier, the data center manager is planning for 10 servers @ 210W each, with expected power consumption of 2.1kW (total), but the servers nameplate power rating is 300W. So, he is allocates 2.5kW for that rack (just in case), but doesn’t know what would be the actual consumption until the servers are all loaded up and running. In the case of busway, we could allocate 100A, 225A or 400A for the entire row of racks in a single busway (and we could have dual busways for dual redundant sources). Within this capacity, it is possible to slice and dice, with whatever excess reserved or allocated power difference from the actual consumption, it just leaves additional capacity for new loads or equipment.
There are many configurations of plug-in units, and the beauty of these plug-in units is that we could have actual metering to measure real-time consumption. Connect this to your intelligent Building Management System and voila! You get real-time consumption data…
The advantages of this method of power distribution over traditional pre-laid power cables are many and includes:
>> Scalability: Enabling components to be added as needed, instead of building out the entire facility in the beginning. Day one requirements are often different from original plan, and therefore, why design fixed electrical cabling and outlet locations?
>> Monitored Power Usage: The ability to monitor and know the exact power usage, at any point in time, within the data center is important for proactive management. All too often, as new equipment are added into a rack, the cable ratings may be exceeded and/or risk of tripping the circuit breakers, causing unplanned outage.
>> Flexible, Adaptable power distribution: Compared to pre-laid power cabling, the busway is versatile (I could join two tracks to have a longer track, or combine them into an L-shape or T-shape busway) and the system can be tapped at any location with a variety of plug-in units, eliminating panel boards, long runs of conduit or wire, and expensive installation costs, etc. Dedicated circuit breakers at the point of use make reconfiguring supplies easier and troubleshooting easier.
For a colocation or hosting service provider, such flexibility and scalability is essential as it will ease a lot of pain and costs in the long-run. You’ll never know when the next customer will be reeling in a rack full of Blade systems, or a huge enterprise class chassis of servers or storage arrays.
Tags: Data Center, Density, Efficiency, electrical