Imagine the scene: a data centre (the Hammersmith Appollo) with a main machine hall (the stalls + stage) and a half-size mezzanine (the circle). 3600 racks (or seats), quite well spaced, filled and with occasional hot-spots . (One or two of the older models must have been liquid-cooled given how frequently they had to visit the facilities:)). It was cooled to a comfortable temperature, not excessively cold. And it was very much a mixed vendor environment. On stage there was
- a compact, high energy server (Ronnie Wood on lead, perhaps the man who put the “64” into x64)
- an old mid-range, running very smoothly and efficiently on background processes (Bill Wyman on bass)
- some high-reliability Unix machines keeping core applications running, rock solid (Mick Taylor and Hamish Stuart on rhythm, Andy Newmark and Simon Kirke on drums)
- some virtual machines on standby to add extra processing during peak periods (the horn section)
- and some multimedia servers for streaming specialist content (singers Beverley Skeete and Sharleen Spiteri)
- not forgetting the amount of archive storage these machines need access to (looooonnnnng memories)!
Overall, it was a solid performance without too many risks being taken. Perhaps the data centre equivalent of a PUE of 2.
Of course, there is something else a concert review has in common with all too many data centres: it is purely subjective. You cannot rely on canvassing a few audience members after a performance to get a true impression of what it was really like. Similarly, many data centres rely on taking some manual readings at perhaps weekly or monthly intervals (or worse), transfer them to a spreadsheet and then calculate a one-off PUE value. This does not tell you anything very useful about how efficiently you are running your facility. To really understand data centre efficiency you need a DCIM solution centred around automated, detailed monitoring of the infrastructure combined with informative reporting. Perhaps something like InSite, for example.