Large web and cloud services companies like Amazon and Microsoft
run some of the most efficient datacentres in the world – so what lessons could
enterprise IT learn from them?
Analyst house Forrester interviewed staff from the 10 major cloud service and hosting providers – including Amazon, Microsoft and salesforce.com – to
find out the practices that allow them to run efficient datacentres at scale.
“These providers have perfected the art of running a
simple and standard environment that is optimized to each specific workload, and
automated as much as possible,” the report Five Data Center And IT
Infrastructure Lessons From The Cloud Giants states.
While acknowledging that IT estates within your
average enterprise are likely to handle a greater mix of computing workloads than your average infrastructure-as-a-service provider and to run these workloads at a vastly smaller scale – the
report identified five practices of the web giants that could be applied to IT operations of
1,000 or more servers.
Build your datacentres around a standardised IT infrastructure
“Start shifting workloads to a small set of standard configurations
that will be automatically deployed,” the report states, stressing that
virtualisation provides the necessary abstraction from the underlying hardware
to make this flexible deployment possible.
Infrastructure complexities, such as synchronous storage replication and live workload migration, should not be hooked directly into applications, it says, but should be utilised below the abstraction level these services sit upon.
Automate the basics
The large web players generally choose to use scripts,
open source tools and homegrown management systems to automate the running of their datacentres as much as possible, the report found.
Tasks suited to automation include essential system configuration – such as setting up system IDs and MAC addresses, and software installation – which the report
says should be carried out using image-based installations wherever possible.
Tools to help with this automation range from commercial products
from the likes of CA, BMC and IBM to open source software like Puppet, Chef and
The largest challenge in automating processes is “the
inability of most I&O (infrastructure and operations) groups to develop
processes that lend themselves to automation”, according to the report.
Change from infrastructure management to service delivery
Try to think of the IT department’s role as less about
looking after the infrastructure and more about delivering the best IT service –
regardless of how and who delivers it.
The IT services provided by IT ops to the rest of the
business and developers should be measured against targets related to the service being delivered, rather than to the running of the infrastructure, and should be competitive with public cloud services in key areas such as “cost-transparency” and “service efficiency”.
IT organisations should be willing to change how a service is delivered – to switch
service providers or commoditise the underlying hardware infrastructure – if it
will improve the efficiency of the service. The freedom to make these changes
with minimal disruption comes from abstracting IT services away from the
underlying hardware, according to the report.
Break down the silos inside IT
Insist on collaboration between the teams looking
after IT facilities, server, storage, network
and virtualisation. For example hosting provider Savvis told Forrester it has an internal
wiki that all of its IT teams use to document “issues or complex
configurations”, as well as a policy stopping groups like ops teams and database
admins from making unilateral decisions on configurations and deployment.
It should also be recognised that DevOps has an increasingly important role within the modern
IT team, as they need to be able to architect software to take full advantage of distributed and standardised nature of cloud platforms.
Automate facilities management
Enterprise can take advantage of the advanced power and cooling management techniques similar to those used by the large web properties, thanks to modern datacentre infrastructure management products and services.
These offerings can drive down operating expenses (opex) by offering tools for managing and granular monitoring of such as Power Usage Effectiveness (what proportion of the datacentre’s power consumption goes towards running the computing hardware), overall power consumption, cooling infrastructure, UPS and battery status.