|
||||
|
|
|
|
|
Server Configuration
When "Green" Isn't Really Green
The goal is to service your computing workload in the required
time while using the least amount of energy. In an ideal scenario,
all factors influencing energy efficiency on the supply side
(data center) and demand side (IT equipment) can be controlled
to most effectively deliver the greatest energy efficiency.
Unfortunately, this is almost never the case, so it is necessary
to make the most appropriate choices using the factors within your
control. It is important to view the efficiency of the complete solution versus efficiency on a component-by-component basis, as the latter may actually result in unintended consequences that can reduce the overall energy efficiency of your computing infrastructure. An example of this is choosing the lowest-power configuration of a given server based on its maximum power usage. This strategy works well when the compute performance of a server is not critical to it servicing its workload, and the server is provisioned with only a single workload. It may not work well when multiple workloads are to be consolidated onto a single server, or when the server is a compute resource for a server farm, database, or compute cluster. In these cases, the overall energy efficiency of more low-power-usage machines often is less than fewer high-performance machines servicing the same workload.
Server Consolidation Through Virtualization
For more on server virtualization, visit the Virtualization section of our website, and see our new Virtualization Platforms.
Know Your Workload
Power Management
Most data centers are sized for workloads that represent demand peaks.
In many cases, the actual IT equipment utilization is much lower than
the peak workload for which the infrastructure was designed. Currently,
most servers operate at approximately 70% of maximum power even when
they are at idle. This means that many servers may be consuming as much
as 80% of their maximum power even though they are operating at only
20% peak processor utilization. Power management features are often
disabled because of concerns regarding response times and server
performance degradation.
Server processors have power management features built in that, when enabled, reduce power at idle. A server with processor power management enabled can reduce power to less than 50% of maximum power at idle. Typical business data center workloads are near their peaks from 5 AM to 11 PM, leaving at least 25% of operational time available as a target for enabling power management settings. |
Provisioning one workload on one server can lead to an underutilization
of resources. This situation is often referred to as "server sprawl",
a condition in which servers consume more resources than is justified
by their workloads. Using virtualization, multiple workloads can be
consolidated onto a single server to reduce the energy required
to service each workload. With enough processor, memory, and storage
capacity, it is not uncommon for a single server to consolidate 10
or more workloads. By allowing hardware consolidation, virtualization
reduces the number of servers to power and manage and enables
power supplies to operate at their peak efficiency. Up to 25% savings
in power consumption is often reported post server consolidation.
Understanding your workload and the factors that contribute to
improving server performance for servicing that workload are
important. Knowing the manner in which changes in processors,
RAM, I/O, and other hardware affect a server's ability to service
your workload allows you to choose the right server, and the right
quantity of servers, to optimize the energy
efficiency of your infrastructure. The industry is developing
standardized workload and power benchmarks that will enable
you to better choose the compute infrastructure that is the most
efficient for your workload.
| previous page | next page |