Thanks Tim for clarifying.
On Wed, Mar 30, 2016 at 3:39 PM, Tim Starling <tstarling(a)wikimedia.org>
On 31/03/16 02:55, Katherine Maher wrote:
IIRC, we included clean energy consumption as a
evaluating in our RFC for our choice of a backup colo a few years back
Since I strongly support emissions reduction, on my own initiative I
did an analysis of expected CO2 emissions of each of the candidate
facilities during the selection process of the backup colo. That's
presumably what you're referring to.
My conclusion was that codfw (the winner) was one of the worst
candidates for CO2 emissions. However, the price they were offering
was so much lower than the other candidates that I could not make a
rational case for removing it as an option. You could buy high-quality
offsets for our total emissions for much less than the price difference.
However, this observation does require us to actually purchase said
offsets, if codfw is to be represented as an ethical choice, and that
was never done.
codfw would not tell us their PUE, apparently because it was a
near-empty facility and so it would have technically been a very large
number. I thought it would be fair to account for marginal emissions
assuming a projected higher occupancy rate and entered 2.9 for them,
following a publication which gave that figure as an industry average.
It's a new facility, but it's not likely that they achieved an
industry-leading PUE since the climate in Dallas is not suitable for
evaporative cooling or "free" cooling.
Ops runs a tight ship, and we're a relatively
small footprint in our
so we don't necessarily have the ability to
drive purchasing decisions
based on scale alone.
I think it's stretching the metaphor to call ops a "tight ship". We
could switch off spare servers in codfw for a substantial power
saving, in exchange for a ~10 minute penalty in failover time. But it
would probably cost a week or two of engineer time to set up suitable
automation for failover and periodic updates.
Or we could have avoided a hot spare colo altogether, with smarter
disaster recovery plans, as I argued at the time. My idea wasn't
popular: Leslie Carr said she would not want to work for an
organisation that adopted the relaxed DR restoration time targets that
I advocated. And of course DR improvements were touted many times as
an effective use of donor funds.
Certainly you have a point about scale. Server hardware has extremely
rudimentary power management -- for example when I checked a couple of
years ago, none of our servers supported suspend-to-RAM, and idle
power usage hardly differed from power usage at typical load. So the
only option for reducing power usage of temporarily unused servers is
powering off, and powering back on via out-of-band management. WMF
presumably has little influence with motherboard suppliers. But we
could at least include power management and efficiency as
consideratons when we evaluate new hardware purchases.
At the time the report came out, we started
talking to Lukas about how we
could improve our efforts at the WMF and across the movement, but we've
limited bandwidth to move this forward in the
Foundation (and some
transitions in our Finance and Operations leadership, who were acting as
executive sponsors). However, I think it's safe to say that we'd like to
continue to reduce our environmental impact, and look forward to the
findings of this effort.
We could at least offset our datacentre power usage, that would be
cheap and effective.
-- Tim Starling
Wikitech-l mailing list
149 New Montgomery Street
San Francisco, CA 94105
+1 (415) 839-6885 ext. 6635
+1 (415) 712 4873