* Data center managers need useful information about electrical power
An earlier newsletter explored the idea of a cabling plant that was just smart enough to describe itself, to human eyes via LCD panels, or to other IT systems via a standardized information feed.
An aspect of data center infrastructure not discussed there is power. Power feeds are another data center resource that has to be managed carefully: which power distribution unit (PDU) on the rack should this server go into? Which feed under the floor can this PDU be plugged into? For many or most data centers this is ultimately managed with spreadsheets and labels applied to power boxes and PDUs.
There are other kinds of information IT can easily get from data networks that are rarely available for power networks. How much power is being pulled right now on this circuit through the main panel, or this one through the UPS, or this under-floor outlet, or this PDU? Right now, how much power is being consumed, where, and how does that change over the course of a day, or a month, or a year?
With information like that in hand, it would be possible to more intelligently manage the placement of servers in racks, or even the flow of virtual servers around a resource pool. One goal might be to avoid hot spots and keep heat load spread out optimally (not evenly, mind you – it’s still better to have some systems power down and others run at higher workload than to run them all at very low workloads).
Another goal might be to see that power-draw/heat-push remains pretty constant in each rack throughout each day. Such information would also remove some guesswork from capacity planning.
IT could obtain this kind of information if, as with the semi-active cabling plant described in that earlier column, the data center’s power distribution system woke up and was able to report on its own status. Presumably, with technologies for power-line data connectivity becoming available, sending information reliably within the current flows themselves should be possible, allowing power network to be information network too.
Leaving aside in-stream data flows, the best place to start would be rack PDUs: a PDU with a Category-5 connection to a data switch in the rack could provide information on how much each feed off the PDU is drawing, at regular intervals or on demand, as an XML stream. It could also report on how much power it is pulling in, allowing data center management software to watch for things like failing PDUs (if efficiency begins to fall) as well as to report on the pull from wall or UPS power circuits.
Working out from there, and possibly utilizing the power stream for communications, intelligence might flow further in, into servers, or further out, to floor feeds, wall panels, and UPSes.
Since many UPS systems are now smart enough to report on what each feed in or out is doing, the big change here would be UPSes getting the ability to read information from or send information to power units in the data center, or upstream to main circuits in the building or elsewhere. Within a server or chassis, power supplies could gain the ability to report on how much power was going to each subsystem in the case or chassis – CPU separate from storage, separate from NIC.
We have seen, through successive generations of IT evolution, that when it comes to managing resources, information is power – the more we know about a resource pool, the better we can manage it. In this case, for resource management and capacity planning purposes, power is information, too.
