56
influxofmoreefficient
serversandgreater
virtualizationand indicates
there isanopportunity toadd
capacityby increasing rack
density.
When rackdensities rose to6
kW in2006, itcreated
problems formany facilities
thatwerestill relyingona
cooling infrastructure that
was installed 10or 15years
earlieranddesigned to
handledensitiesof 1 to3kW
per rack.Cooling
technologieshaveevolved
considerablysince then.
Perimetercoolingsystems
nowoperateatmuchhigher
levelsofefficiencyandhave
built-in intelligence that
allows them tocommunicate
andcollaborate. Inaddition, coolinghas
steadilymigratedcloser to thesource
ofheat, increasingefficiencyand the
ability tocool higherdensity racks.
Aisle-and rack-basedsystemscan
safelysupport20or30kW racks.With
theproperpowerandcooling
infrastructure inplace,mostdata
centerscandoubleor triple their
existingcapacitywithout increasing
datacenter space. Thiscanalso
enhancedatacenterefficiency, as
denserdatacenterenvironmentsare
inherentlymoreefficient than
environments that spreadout the load.
Ifadditional capacity is required, an
aisle-basedorcontainer-based
expansionstrategycanbeemployed in
which initial capacity ismetby the
requirednumberofaislesorcontainers
but spaceandpowercapacityare
reserved for theadditionof future
“modules.”Whencapacity isneeded,
additional aislesorcontainers—with
integratedcooling,monitoringand
powerprotectionanddistribution—can
beadded, enablinganeasy-to-
implementmodulargrowthstrategy
(Figure5)
.
Thisapproachhasbeenespecially
popularwithorganizations thatneed to
expandquickly to react tomarket
demandsoropportunity, suchas
colocationproviders, or those
deliveringcloudservices.Withproper
planning, significantblocksofnew
capacitycanbeadded ina fractionof
the time itwould take toconducta
traditionalbuild-outorbuildanewdata
center.Because theyhave theability to
respondquickly, theseorganizations
can reduce theirupfrontcapital costs
and increaseoperatingefficiencyby
usingahigherpercentageof their
operatingcapacityat startup.
Alternately, capacitycanbeadded
throughcloudorcolocationproviders,
buteven in thiscase, infrastructure
remainsan importantconsideration.
First, the in-housedatacenter
infrastructureshouldbeable toadapt
tovarying loadswithoutcompromising
efficiency.Perhapsmore importantly,
the infrastructureof thecloudor
colocationprovider shouldbe
evaluated toensure ituses
technologiesandconfigurationsproven
tosupporthighavailability.
PROTECTINGAVAILABILITY
While it isdifficult topredictexactly
whatwillbeexpected from thedata
centerof the future, it ishard to imagine
ascenario inwhichdowntime isn’ta
serious issue.Datacentersattempting
tocompletelyeliminatepower-related
downtimegenerallyusedual-bus
architecture toeliminatesinglepoints
of failureacross theentirepower
distributionsystem. Thisapproach
includes twoormore independentUPS
systemseachcapableofcarrying the
entire loadwithNcapacityafterany
single failurewithin theelectrical
infrastructure. This isaproven
approach fordeliveringTier IV
availability, butdoes requirecustom
switchgearand limitspowerequipment
utilization to50%, impacting initial
costsandoperatingcosts.
Alternateconfigurationshaveemerged
in recentyears tosupporthigh
availabilitywhile increasingpower
equipmentutilization, including
distributed reservedual-bus
architecture. Static transfer switches
(STS)areused toprovide redundancy
acrossmultipleUPSsystemsaswell as
the transfer switch itself
(Figure6)
.
Asimplerversionof thisarchitecture is
representedby the reserve-catcher
dualbus. Thisapproach isextremely
attractive toorganizationsseeking
dual-busavailabilitywith lower initial
costsandgreaterefficiency. Like the
7X24MAGAZINE SPRING2014
Figure5
Figure6
