FOCUS CLOUD INFRASTRUCTURE
Issue 10, June/July
OF DATA CENTER DEMAND RESPONSE Few companies have the appropriate power infrastructure and ‘smart grid’ still has a long way to go
them to reduce cost. A group of scientists in California suggests that big companies with extensive geographically dispersed data center topologies that act as private clouds should also use these capabilities to participate in demand response, which will help reduce cost further and more.
T
Organizations that leverage their clouds to shed load when grid operators ask them to will make substantial contributions to efforts to lower strain on power grids that serve their data centers and other businesses and residents in their regions. A cloud-like infrastructure is key. This is suggested by the latest study on demand-response opportunities for data centers conducted by the Department of Energy’s Lawrence Berkeley National Laboratory in California, in collaboration with K.C. Mares of Megawatt Consulting and David Shroyer of Shroyer Consulting Group. One of the study’s conclusions is that virtualization – prerequisite of the cloud delivery model – is the biggest opportunity for demand response in data centers, providing the ability to reduce both IT and cooling power loads.
The idea is simple: when one data center receives a message from operator of the electrical grid it is connected to that load on that grid is high at the moment, the data center drops its electricity consumption by transferring a portion of compute workloads it is performing to a facility in a different location, where load on the local grid is lower. According to Girish Ghatikar, business and systems analyst at LBNL who headed the team behind the study, said some data centers already have the infrastructure necessary to implement such a strategy. Today, these capabilities are primarily used for disaster-recovery purposes and to cut cost. While implementation of demand- response strategies will cut cost further (power is more expensive at peak demand), all users also have the responsibility to do their part in prevention of brownouts and blackouts, in
24
www.datacenterdynamics.com
he case most often made for cloud computing is that it increases efficiency of the way companies consume IT services, enabling
CLOUD INFRASTRUCTURE KEY ENABLER
that are easily susceptible to virtualization and that can either reduce power consumption as their compute load drops or quickly switch themselves off or go to a low-power mode when idle and then, just as quickly, get back up to normal processing capacity when needed.
Ghatikar’s opinion.
EXISTING OBSTACLES Main obstacles to a widespread implementation of such strategies today are the fact that very few companies have the infrastructure necessary to shift loads in this way and the fact that few utilities offer real-time pricing for energy at retail level – a common practice in the wholesale electricity market.
Successful implementation of demand response by data centers on a large scale would also require comprehensive real-time assessment of power availability in various geographic regions and demand response that is coordinated among all users and power distribution grids involved. Such a level of coordination is needed in order to avoid destabilizing portions of the grid that load is shifted to. One argument against investing into smart grid and focusing more on on-site co-generation is that electrical infrastructure in the US is old and overloaded. In Ghatikar’s opinion, successful coordinated demand-response events can help reduce strain on the grid and reduce additional investment of public funds into building out more electrical infrastructure.
DATA CENTER INFRASTRUCTURE PREREQUISITES
In addition to a smart grid outside the data center, the mission-critical facility capable of working with the grid to balance out demand has a set of necessary characteristics. First one, as already mentioned, is a cloud-like IT service delivery, where the infrastructure is shared by applications neither of which is reliant on one particular set of hardware to perform adequately.
This delivery model requires advanced servers
One reference LBNL makes is to a 2008 proof- of-concept project conducted by the storage vendor NetApp at one of its data centers to study consolidation by virtualization. While keeping details of the study private, NetApp concluded that servers can potentially be idle for up to 26 percent of the time. Another conclusion was that virtualization policies could be used for “graceful power-on and power-off.”
Also critical to demand response by compute- load shifting is extremely robust network infrastructure that connects the data centers involved, as the strategy would enjoy very little popularity if shifting compute loads between sites was not seamless. Such network infrastructure is not yet available in the US at mass scale.
Finally, automation is crucial in the process. “You can’t do this manually,” Ghatikar said. “No way. You need to have an intelligent technology infrastructure to be able to do that.”
DEMAND RESPONSE POSSIBLE TODAY
Even though there are still significant barriers to successful wide-scale demand-response participation, select companies can and do participate in demand-response events even today. Utilities, like PG&E in California, offer advanced notification for such events and companies, if they so desire, can prepare to shift their compute loads when such events are scheduled to happen. When designing a system that will allow them to participate in demand-response events, Ghatikar recommends that companies consider two dimensions: depth (how many hours can they drop the load for at a particular location?) and breadth (how many kilowatts they can drop load by?). “Those are the things that will help you to build the platform to be able to participate.”
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56 |
Page 57 |
Page 58 |
Page 59 |
Page 60 |
Page 61 |
Page 62 |
Page 63 |
Page 64 |
Page 65 |
Page 66 |
Page 67 |
Page 68 |
Page 69 |
Page 70 |
Page 71 |
Page 72 |
Page 73 |
Page 74 |
Page 75 |
Page 76 |
Page 77 |
Page 78 |
Page 79 |
Page 80 |
Page 81 |
Page 82 |
Page 83 |
Page 84