New Daedalus

I was part of a round-table on Demand-Response in data centers last week. Demand-Response is the strategy used by the power companies to trim peak load, the load that degrades into brown outs on a hot summer day. Demand-Response can include options from forward agreements to get a better price to live bidding to give up power, depending on the regulatory and market environment.

Missing from much of the discussion it was any real focus on marshalling and allocating all the resources available to solve business problems. As a government employee, I felt right at home: cost, cost, cost, when the discussion should have been value, value, value.

Modern management practices in the data center are concerned with optimizing the conversion of compute cycles into raw business process. In the world of blades and virtual machines (VM), concerns with servers are replaced with business process throughput, and trading schedules to improve service levels.

Concomitant with this is the most efficient direct conversion of electricity to heat ever devised. Heat disposal becomes paramount. Support infrastructure, whether additional cooling capacity or electrical capacity is rarely managed at all. Corporate resources are thrown away because excess energy is treated as a waste stream.

The best data center management is always trading off business resources to meet business service level agreements (SLAs). If transactions through the middle tier supporting sales transaction take longer than, say, three seconds, customer will begin to abandon transactions. Using blade servers, a new VM is quickly provisioned to supply the needed compute power. If there is no payroll this week, transactions supporting human resources can be delayed, a VM shutdown, and a new VM provisioned for sales transactions. The modern data center is managed for business process delivery, not for computation.

Under this model, Demand-Response is simply a new business cost, a new service trade-off. Demand-Response is simply a further management of the same business process in the same way as everything else. Based upon cost, some business processes are slowed down, or even transferred to another data center in another part of the grid.

I guess what I am arguing is that if you manage your data center properly, Demand-Response is just another business event, and not a unique one. Oh, and make sure your back-up data center is on a different grid.

Data centers are struggling to shed their ever growing heat loads. They are coming up with more and more creative ways to manage energy use, to interact with building air conditioning, to somehow solve the huge load they place on all the resources of a building. I think that some of the proposals out there right now will sound like steam punk in ten years.

Steam Punk is a genre of science fiction that deals with alternate histories constructed around some small change in discovery preserving a now outmoded technology. What if that small note that was discarded as improbable had been recognized, and developed? What if the Greeks had recognized that their steam toys could do work, and thereby been able to defeat the Romans? What if Archimedes had come up with some of his many inventions at any time other than the siege of Syracuse, launching an early industrial revolution? What if the Chinese had dedicate military explosions to ballistics rather than pyrotechnic shock and awe?

The most popular forms of steam punk present neo-Victorian or neo-Edwardian realities. What if Ada Lovelace’s working computers had been translated into working production system after a chance meeting with James Watt? What if the French had developed the Jacquard loom into an alternate computing engine? What if the Napoleonic wars were thereafter fought using steam-driven military computers as steam-engines drive military dirigibles through the skies?

Zero Net Energy buildings are up and coming. The common core the Zero Net Energy buildings is local storage and conversion of energy. Night-time energy prices are stored in ice for daytime use. Daytime solar energy uses molten salt to buffer electrical generation. Perhaps windmills will pump water to roof-top cisterns for emergency electrical generation. Energy storage and conversion are central to every strategy for near-grid buildings.

The astonishing heat of the data center is an energy source that should be tapped, not dissipated. Data centers can heat office space in winter. If we concentrate rather than diffuse machine heat, we can cool office space, in part, using absorption chillers drive by heat. Domestic hot water could cost the landlord the same as cold water. Shedding heat load simply makes no sense in the zero net energy building.

If a landlord can get a data center paying rent in the basement, while harvesting the heat load to drive building operations, then he is doubly rewarded. The technology is rather straight forward. Understanding the tenant contracts and incentives is something else.

The data center can be the steam plant of tomorrow. Their heat is a resource to be harvested. This will make many of the careful technologies of The Green Grid seem almost quaint. Like the technologies of steam punk.

Normal business now need to defend themselves from the power systems just as data centers do.

Long time readers know that I consider that power companies and utilities commissions over-estimate power reliability by focusing on the presence of power and not on the quality of power. I often note that my home, halfway between a major research university and a nuclear plant suffers multiple outages a month, outages long enough to require that I reset all the devices in the house, whether microwave, DVD player, or alarm clock. This probably has something to do with the frequency with which I must replace home electronics. Yet homes are “adequately supplied” with power.

On the other side is the data center. Data centers have long acknowledged that utility power is neither good enough nor reliable enough for their purposes. Data centers use multiple strategies for on-site energy storage, on-site energy generation, and on-site energy conditioning to protect themselves from the product supplied over the power grid.

Non high-tech businesses are considered as something more similar to the home than to the data center. They were not worth protecting in the way data centers are protected.

Two weeks ago, some friends open up a bakery and sweet shop in Chapel Hill. Sugarland is an all-natural bakery and gelato shop. Its business equipment is kitchen equipment and retail refrigeration. It seemed the worst problem they were going to have was keeping up with the swarms of students that found them as soon as they opened, before their staff was all trained. Doc and Katrina were exhausted, but pleased. The snacks were delicious. Rush hour warm cookie time was a success.

Last Sunday a wind storm came through the southeast. In town power would flicker, then flicker again. One would think that this business would be mostly unaffected, not much different from the businesses the grid was designed for in the 1950’s.

Modern ovens, however, have computer systems the run them, computers that reset with each flicker. Modern gelato machines have processors that stop when the power dims a little. The cash register is, of course, a high touch system for inventory control and minimal staff training, until its database corrupts.

I went by Sugarland on Sunday as Katrina threw out 250 cupcakes that deflated when the oven re-set. She did not dare start more cakes for the morning until she knew the power would be reliable. Doc had given up on trying getting fresh gelato out to the waiting lines. Neither knew what to prep that night to prepare for the early morning baking on Monday.

The absolute shut-down and loss of business for flickering power in a modern retail bakery is as big a hit as in any data center. Bakers, too, need to defend themselves from what comes over the power line with

Monday’s short stock is now over. The power has been adequate this week. The shelves are stocked again. But I will no longer consider data centers as having special needs; merely needs that are better recognized.

I think I will head downtown now – I hear the grapefruit gelato is superb.

Earlier this week, I suggested that one outcome of the Software as a Service (SaaS) approach will be that complex operations such as BIM servers would migrate into cloud computing, where there data can be shared and updated by Designers, Architects, Engineers, and used by Owner/Operators.

This week, Sun (the computer company that long used the motto “the network is the computer”) announced that by 2015 Sun will have *no* data centers. All internal IT for this technology company will be acquired under the SaaS model. "We will need to get to a point in which we mandate detailed SLAs and manage/monitor those SLAs," Cinque writes. "As long as a SaaS provider can adhere to our detailed SLAs, then it shouldn't matter where the applications sit. The challenge is getting those detailed SLAs written out, (and) having the SaaS industry evolve where they can accept client-driven SLA's." By 2013, 5 years from now, Sun expects to eliminate half of their square footage in data centers

So what does this say about buildings and energy?

First, it both intensifies the reasons behind The Green Grid (see my earlier comments here) while reducing the number of people interested. While CABA may call Networking the 4^th utility, data processing may instead become the 4^th utility. The differences in cost, and the intensity of the operations load may make data centers an unjustifiable expense.

Rapidly rising energy costs may be the catalyst; virtualization and the associated capability of outsourced customization to produce tailor-fit solutions are the enabler. The increasing demands of security and enhanced benefits of service specialization will be long term drivers. Those data centers that remain will demand highly efficient interactive means to manage transforming energy to raw business process.

Next, it invalidates the traditional building system model of on-site monitoring. Traditional building systems bring too much data and too little information to too few people. Cloud computing will decrease the barriers to fewer people with greater expertise monitoring more building systems. Communications models for embedded building systems that do not match the new reality will suffer competitive disadvantage.

Low level protocols will remain much as they are; gateways that abstract underlying system up to actionable alarms and descriptive information. These gateways will use enterprise-style protocols that perform well over the internet.

Such gateways are exactly what we need to enable new business service. Remote analytics. Knowledge based maintenance management. Live energy modeling. Third party interactive demand/response. Each of these services, and others we do not yet know, will also become services, and move into the clouds.

http://www.datacenterknowledge.com/archives/2008/Jan/10/suns_goal_no_in-house_data_centers_by_2015.html

My early conversations at Grid-Interop this week have been driven by The Green Grid, and the discussions they have been having.

The Green Grid does not refer to the power grid, but to grid computing. In essence, the Green Grid is trying to solve the problems of reliability and efficiency in data centers. Data centers consume large amounts of power and convert it business process and heat. Green Grid operators want to understand the reliability of their power source, they want to know how well the building systems will be able to dissipate the heat, but the only thing they want to manage is the business processes.

Last night I had dinner with Ken Uhlman. Ken has been working on the Green Grid for the last year or so, and is passionate when describing the power distribution and power quality management. Ken has seen many of the same problems I have while working with oBIX. Many of the best engineers are detail oriented and love their work. They love their work so much they want to share it with others, They want to share not only how well their systems work, but all of the details about how they work.

Extraneous details are the enemies of interoperability. Extraneous details vary from implementation to implementation. Extraneous details lead to lock-in.

More importantly, extraneous details increase complexity. Complex systems are difficult for people from a different domain to use. If the complexity is too daunting, people just won’t try it.

So Ken and I were discussing a variety of issues when we came up to the crux. There are five elements of information that the data center needs from its electrical systems.

• What’s my current burn rate, i.e., how much energy am I using NOW?
• What’s my current price?
• What quality of power am I receiving now (may be a complex data type)
• What is the probability of some sort of failure now and for the near future?
• ( variant of 2) What predictions for price do you have in the near future?

With that information, and no others, the Green Grid would get the information that it needs.

For internal power systems, (breakers, UPS, generators) he suggested one more feature, for whether a service order has been acknowledged after a fault. Possibly a similar feature with estimated repair time could be added to the external feed—but I wouldn’t wait to get it. The first 5 are enough.

One of my visions is a building agent, enterprise aware, choreographing building systems, and getting input from the grid. The inputs Ken has described would be all I need to come from the grid. With that information, and no more, buildings could be responsive to conditions on the power grid.

See, I don’t want much.