nanog mailing list archives

RE: rack power question


From: "Alex Rubenstein" <alex () corp nac net>
Date: Sun, 23 Mar 2008 20:53:05 -0400


A very interesting thread...  I believe that the cost of power will
continue to climb well into the future and that it would be foolish to
build any new infrastructure without incorporating the ability to pay
for what you use.  So this means being able to measure the power
consumption for each server or to aggregate it to the individual
customer.  Then factor in your cost of providing that power and the
associated cooling loads, add you margin and bill the user
accordingly.  Paying for what you use is inherently fair - and I think
of the colo provider also as a technically competent provider of
"clean", highly-reliable power.

Agreed. About four years ago, we saw the writing on the wall. The days
of $750 racks which includes 20 amps of 120v are long gone. It was fine
when the customer consumed 2 amps. No one really thought about the cross
subsidization issues.

Today, we sell racks for $500, but include no energy; every outlet to
every rack in our datacenter is metered, and we sell energy at $n/amp.
It's that simple. In effect, you can have as many outlets as you like
(one time charge for them), but you pay for the consumption.

No question cost of energy is going up and up; we've seen a 75% to 100%
increase in 5 years in Northern NJ. This is why we plan on investing
heavily in solar in the next year or so. With returns being as fast at 6
to 8 years, you are crazy not to look into this.


If you don't have a pay-as-you-go billing model, then what incentives
are there for the users to consolidate apps onto fewer boxes or to
enable the power saving features of the box (or operating system) -
which are becoming more widely available over time?  Answer: none -
and human nature will simply be lazy about power saving.

Even so, this doesn't seem to factor in. Customers have computing needs.
Rarely do you see a customer say, "wow, this amp is costing me $n, so I
am not going to run this SQL server."



Current thread: