Wednesday, June 25, 2014

Sap Data Center Power

'Conaatose'Servers


Decommissioning servers is a difficult
process, as Barclays learned in its nowe
successful effort to rid itself of underused
systems. By Patrick Thibodeau

..............•..........................•.................
A
N ENVIRONMENTAL PROTECTION AGENCY plan to
cut carbon dioxide pollution
, in part through ef-
ficiency improvements, could put pressure on data
centers with servers that are doing very little work
- or none at all.

Many high-profile data centers rurl by the likes of Apple, eBay
and Google incorporate alternative energy into their power mi
x,
and all boast about the efficiency of their op-

erations. But a recent Uptime Institute survey
suggests that a large number of data centers
"are running substantial numbers of servers
that do nothing.

Nearly 25% of the more than 1,000 enter-
prise data center operators and executives who
responded to the Uptime survey said that at
least
10% of their servers are "likely comatose."
And the actual numbers may be higher.
"Most data center operators can't even tell
you how many servers they have, never mind
their utilization, so caution in interpreting

THINKSTOCK

those numbers is indicated," said
Jonathan Koomey, a research
fellow at the Steyer-Taylor Center
for Energy Policy and Finance


at Stanford University. "The
percentages for comatose servers
are likely much bigger
."



Managing underused servers
and improving efficiency isn't
easy, as Bardays has discovered.

Last year, Bardays decom-
missioned about
9,100 physical
servers, which represented
12% to
17% of the financial services com-
pany's total server footprint
. Those
systems collectively consumed


2.5 megawatts, and Bardays
became a model for the industry
when it took them out of service.




That success was the end
result of a multiyear effort to
develop a decommissioning
process.


That program began after
Bardays took on a number of re-
dundant systems in its
2009 ac-
quisit
ion of the North American
operations of Lehman Brothers
.
Initially, the decommission-

ing process was chaotic. "What
we learned was the biggest impediment to success was people
's
reluctance to click the 'approve' button" on a change ticket, said
Paul Nally, a director at Bardays.


Shutting down a database server might involve a dozen tickets,
but there was no order to the process
. The Bardays IT operation
used an orchestration software tool to ensure that the approach
followed a series of orderly steps: Database administrators went
first, followed by storage administrators, then operating system
specialists and finally the people who did the physical work
.

To overcome concerns about decommissioning, Nally said the
process was made as "safe as possible" and "reversible." With
those controls in place, he noted, "people's reluctance to hit
'approve' sort of abated.
"

Shutdowns take place over three weekends. On the first
weekend, there
's an inventory check to make sure the server is
where it's believed to be; a week later, the server is shut down
via an automated workflow; and the server is
removed from the rack on the third weekend.




The three-week process allows Bardays IT
teams to quickly recover a system if a mistake
is made during the decommissioning.

There were some errors early on, but the
process has improved with the adoption of a
consistent
, repeatable plan.

Decommissioning ser~ers is "cleaning up
after yourself," and in doing so
"you remove a
lot of noise from the environment
," said Nally.
The end result, he added, is a more nimble and
efficient IT opera
tion. +
COMPUTERWORLD.COM 7

No comments:

Post a Comment