A couple of weeks ago I raised the question about hold-up time and the apparent retrograde position that DELL was taking over the 1997 version of the ITIC/CBEMA power curve.  They had, I had been told, declared that their kit did not always meet the 20ms zero-volts specification but was ‘OK’ with the pre-97 10ms specification.  This hadn’t been the first time I had queried the long standing ITIC curve having previously suggested that, surely, power supply technology had moved on?  I’m not sure it was clear from my latest blog that the reason I was raising the topic is not theoretical but very practical since I was considering the wide-adoption potential of eco-mode UPS – so the transfer time needs to be compared with the load immunity.

As some of you will know I post my blog on some of the LinkedIn forums and the previous blog generated almost no traffic which surprised even me.   This time, I suspect because I named a supplier, a lot of traffic ensued and it is still rumbling on – and very enlightening it is too:

DELL themselves confirmed my stated assumption that they didn’t use inferior power supplies.  As nearly all power supplies come out of a (very) few factories in the Far East that raises the interesting point that would every other OEM also struggle to meet the 20ms hurdle?  What DELL didn’t respond to was the challenge that their reticence was due to warranty-avoidance rather than a real condition, but, given their 2011 declarations on server inlet temperature (where they announced that their warranty covered 40C, way in advance of ASHRAE ‘recommended’), I can’t see them being so conservative on power immunity just to be on-the-safe-side.

One of the forums most active participants (and a highly respected US engineer) asked of DELL why they hadn’t said anything before – as they had been members of ITIC for decades.  I’m not sure that their answer was much more than a nervous clearing of the throat and two-step shuffle to the side.

A couple of points remain unanswered for me:

The relationship between hold-up time (stored energy) and server load is not enumerated, i.e. they do have more than 10ms at partial load, maybe even 20ms, but don’t say how you could tell.  So that is like a lottery where you are guaranteed to win but just don’t know how much more than getting your stake back.

No one took the bait on the 120V/60Hz versus 230V/50Hz ‘difference’ – if indeed there is one.  The interesting point is the 10ms ‘old’ spec and 20ms ‘new’ are easy to explain for 50Hz as a half-cycle and full-cycle but for 60Hz (with 16ms per cycle) the 10/20ms are partial cycles.  Interesting, isn’t it?

Anyway, to my original purpose, in any event 10ms is longer than 4ms (the typical eco-mode transfer-time) but the safety margin may be shorter than you expected from ITIC/CBEMA and if you want to get the margin back then GE have patented a 2ms transfer-time machine.  Now that is fast…