Designing for Demand
The days of a nice flat line static load in the data center are over!
It really wasn't that long ago at all that everything we did in the data center was completely insensitive to almost any change in user demand or other factors such as external temperature.
The facilities guys used to treat IT as big resistors. We installed them, turned them on and then they sat there in a rack for the rest of their lives pumping out heat at an almost constant rate, twenty four hours a day, 365 days a year!
Now since we realised this was the most incredibly inefficient way to run IT services and platform things have changed somewhat. But in those days we had fixed speed drives in the CRAC units, we had chillers that ran 24/7, we had IT servers sitting idle and still drawing +80% of their rated peak load. What a waste of money!
Today we are getting smarter and realising that thanks mainly to the ever increasing cost of energy we need to do better. We've dusted off the old idea of virtualisation and modernised it. We've realised that power management on servers is a probably a good idea and that making servers, storage and network equipment much more load-linear is a bright thing to do.
In addition to that the data center has evolved. We've realised that for a large chunk of the world we can use free cooling and turn the chillers off. We can contain hot or cold air in the data center and move to variable speed drives in CRAC or CRAH's (Facilities folks call these VFD's or variable frequency drives as that's generally how their speed is controlled).
Even the software guys are moving load around now and suddenly hundreds of kilowatts can move overnight from one data center to another!
Well, this really causes a headache for data center facilities guys as mechanical and electrical systems are much slower to respond than computers, storage and networks.
Additionally it makes trend analysis very difficult as the baseline could change dramatically overnight. Demand is now dynamic and we need to now deal with a ever-changing environment in the data center. The system needs to be treated as a whole and predictively modeled to comprehend what's going on and how to keep one step ahead.