[NCLUG] Re: "Green" power -- off grid

[John L. Bass]

DJ Eshelman wrote:
> Along these lines- I recently have been approached by two of my 
> clients wondering how they can 'green up' their datacenters.
>
> I'd love to recommend wind turbines, but does anyone know if that can 
> be done in the City limits?
> Solar is of course an option but I've never felt the output is enough 
> to justify the cost.

Depends greatly on "where" in the city limits, as you need significant 
tower height above all up wind objects to get into clean air that is 
free of turbulence. HAWT high efficiency wind turbines that have 
significant output are not cheap, and their output and life expectancy 
are greatly dependent on clean air streams. If you can not get this 
height, then a good VAWT solution may be a better long term alternative 
(there is considerable controversy about VAWT designs inside the wind 
power community, YMMV). Neither are a slam dunk inside cities with dirty 
air flow. Both require considerable thought as to long term return on 
investment.

Solar is actually a great option ... but you have to do the numbers over 
the 20-30 year life of the system, and that has to be paired with a 
storage option like net metering for any 24x7 facility.

>
> I personally get 100% of my power from the city's Wind program- as far 
> as I know that power is coming mostly from Wyoming.  I'm sure Xcel has 
> the same kind of thing going on.  It adds a bit to the bill, but I 
> feel it's exercising my consumer advocacy by doing it- the more people 
> demand wind power, the more turbines will be built and maintained.  
> The more turbines are built, the easier it will be to lower the cost 
> (supply and demand).

So your lights go off when the wind stops? I admit that is rare in Wyoming.

>
> For now my answer to those wanting to 'green up' their datacenters is 
> to utilize Xen and VMWare, Citrix and technologies like that; but 
> ultimately I think (as someone else mentioned before) - cooling said 
> servers is actually the majority of the expense.  Unfortunately, 
> servers are more efficient at creating heat than air conditioners are 
> at removing it.  It's why I'm really looking forward to IBM's research 
> in this field http://www-03.ibm.com/press/us/en/pressrelease/23826.wss
>

Step one, is to cut energy use where possible. That translates into 
doing more with less.

Audit all servers for energy use, and actual required work load. Is the 
lowest power server possible being used? Is the server powered 24x7 when 
it's only used a few hours per day?

Consolidate the work load for all 24x7 applications into as few very low 
power base load servers as possible. Identify peak load, or time 
dependent, applications and tie them to peak load, or time dependent, 
servers that only are in operation during those periods when the low 
power base load servers can not meet reasonable service latencies. Use 
dynamic load sharing between your base and peak load servers to hide the 
variable capacity requirements.

In most organizations this has several benefits besides matching 
available server capacity to demand. First, it will dramatically drop 
the data centers average power and cooling costs. Second, it creates a 
data center that has redundancy and resilience as a core part of its 
basic design, greatly failure proofing it. Third, it allows distributing 
the servers physically which can be an aid in both disaster planning and 
cooling.

As you note, Xen and VMWare are tools that allow server consolidation.

Step two, is looking at server room technologies that improve on air 
handling efficiency and cooling costs.

It is very expensive to move lots of air for cooling ... M*(V^2) 
problem. Careful air handling can reduce the volume and velocity 
required in cooling by air, thus cut air moving costs and the size of 
the machine room required. Consider rack deployment options with 
integrated heat exchangers, or deployment of multiple heat exchanges 
next to server racks with confined air path ducting.

Consider augmenting heat pumps, with outside ambient air for cooler 
parts of the year. Many data centers are designed using only heat pumps 
and water chillers for thermal management. Smaller and lower power data 
centers can frequently be operated in dry areas using external air for 
cooling. Frequently this requires designing the machine room to operate 
over a wider temp range. It can be one natural side effect of replacing 
unnecessaary

By focusing on very low power server technologies nearly all of this can 
be avoided, except where high power computational servers are required.

Step three, carefully audit the work load on each base load server.

Are there tasks that should be moved to specific client machines?

Are there tasks that are not required by the business?