Technology Deployment

Solutions
Datacenter Challenges

“Green Datacenter Through Collaborative

Agenda

 Trends & Issues

 Datacenter Design
 TDS

2
Trends & Issues

 Trends & Issues

 Datacenter Design
 TDS

3
 Trends & Issues - Increasing Density

 Density of computing equipment has been

increasing (i.e., smaller circuits, multiple cores,
blade servers, etc.)
 Old server: 3-4 U, 100 W
 New server: 1 U, 250-500 W
 Decade ago 500-1000 W per rack was typical
 Currently, 42 U rack can easily draw 10-20 kW

4
 Trends & Issues - Increasing Heat
 Rack drawing 10-20 kW can generate 30,000-
70,000 British Thermal Units (BTUs) of heat
 1 BTU is energy required to heat 1 pound of water
1 degree
 A chiller pumps coolant to multiple computer
room air conditioners (CRACs)
 Frequently, the capacity of a cooling device is
measured in terms of tons. 1 ton of cooling
can dissipate ~12,000 BTUs of heat
5
 Trends & Issues - Requirements
Data Center Power Use
 Every kW used by a
server requires
another kW for
IT Equipment
cooling, airflow, Cooling, 25%
Air Movement,
12%

lighting Electricity
transformation
Cooling

Air Movement
 Cooling requirements UPS, 10%
Lighting, 3%
are approximately 3-6 Electricity
transformation UPS

tons per rack IT Equipment,

50%
Lighting

Source: Intel Corporation, 2006

6
 Trends & Issues - Example

 Consider a 1000 sq. foot datacenter with ~10

racks
 10 racks x 10 kW / rack = 100 kW required for
servers (100W per sq. ft.)
 10 racks x 36,000 BTU / rack = 360,000 BTUs of
heat generated
 Uses 100 kW x 2 = 200 kW of electricity
 Requires 300,000 BTU ÷ 12,000 = 25 ton chiller
 Validate 10 racks x 3 tons = 30 ton chiller AND 30
=~ 25
7
 Trends & Issues - Major Cost
 Power and
cooling
requirement can
be #1 and #2 cost
for operating a
data center
 "In some cases,
power costs
account for 40-
50% of the total
data center
operation budget"
(Intel, 2008)

8
 Trends & Issues - Limiting Factor
 Power & cooling are a major issue!
 Power & cooling may not only be a cost issue; it
can be the limiting factor to new deployments
 Datacenter was designed for ~2 kW / rack, but now
requires 20 kW / rack + additional 20 kW
 Power Distribution Units (PDUs) may not support
required throughput
 Even more difficult is the issue of insufficient
cooling capacity. Typically, facility management
would add CRAC units, but this may be insufficient
9
Trends & Issues - Q&A

 How big is your data center?

 Racks
 Power requirements
 Factor cooling & other draws

 Is cooling a manageable cost or is it a

limiting factor?

10
Datacenter Design

 Trends & Issues

 Datacenter Design
 TDS

11
Datacenter Design - Classic
 ANSI\TIA-942 describes classic design
 Specification describes basic hot aisle\cold
aisle configuration
 Server fronts face each other, while backends
vent to create a common hot aisle
 CRAC units intake hot air (along top of
perimeter) and returns cool air (bottom of
raised floor)
 Raised floor acts as plenum containing cold
air which is forced up through floor openings
12
Datacenter Design - Classic

13
Datacenter Design - Classic

Top View - Hot & Cold Aisles

14
Datacenter Design - Classic

Thermal Side View - Hot & Cold Aisles

15
 Datacenter Design - Classic Reviewed

 Classic datacenter design highly effective when properly

implemented
 Biggest challenge is proper path allowing cold air to reach
all devices
 Cold air may not reach devices at the very top of the
racks
 The raised floor must be sufficiently high to
accomodate cabling while providing sufficient plenum
and avoid disrupting airflow
 Even proper classic design may provide insufficient
cooling for some high-density devices
16
 Datacenter Design - Alternate Approach

 Some debate exists around cooling systems without raised

floors
 Overhead cable trays and duct work are used in some
facilities (especially smaller data centers), but management
of air flow is extremely difficult in such a system
 Depends significantly on fans to direct air down
 If classic design cannot remove sufficient heat, an
overhead installations is unlikely to provide better results,
and a local rack cooling solution may be necessary for the
specific hot spot.

17
Datacenter Design - Q&A

 What is the local facility configuration?

 Are raised floors being used? What is their
height?
 Is a hot-aisle\cold-aisle configuration
used?

18
TDS

 Trends & Issues

 Datacenter Design & Solutions
 TDS

19
TDS - Design, Build, Assess

 Good classic design is important

foundation. TDS provides design and
assessment (i.e., model Computational
Fluid Dynamics) to ensure adequate airflow

20
TDS - CFD Remediation

 Ensure optimal airflow & velocity

 Adequate airflow
 Additional blowers
 Adequate CRAC capacities
 Appropriately spaced
 Add blocking panels
 Remove any obstacles (e.g., cable management)
 Space configuration of CRACs\racks
 Well-sealed floor panels & unwanted openings
 Correct orientation of racks
 Sufficiently large cold air return path (i.e., plenum)

21
TDS - Limit to Classic Design

 Unfortunately even correct classic datacenter

design has limitations
 Can require “pushing” too large a volume of cold
air too far
 Energy involved in “pushing” air becomes
prohibitive (like pushing a rope)
 Alternate techniques include local rack cooling
equipment, hot aisle containment systems, etc.

22
TDS - Waste Reduction
 However, best approach is not efficient
waste-handling but waste elimination
(remember multiplying effect of cooling
on initial computer draw)

 Ultimate goal is to reduce client’s energy

cost

23
 TDS - Waste Reduction

 Develop Service Catalog \ App. inventory

 Identify availability requirements: standard business,
extended business, and high availability
 Reduce & consolidate servers
 Use reduced power modes during non-SLA window
 Virtual server technology
 Performance of new hardware may enable
consolidation (e.g., multi-core servers)

24
 TDS - Utilization
 Equipment provisioned to handle peak loads
 Since peak load is transient, systems are left with low
utilization remaining time: > 20% average utilization
 While demand fluctuates throughout the week, data
center power demands are generally flat
Mainframe CPU use by shift

80.00%
67.90%
70.00%
57.60%
60.00%

50.00%

Utilization
37.70%
40.00% Series1

30.00%

20.00%

10.00%

0.00%
Prime Shift Non-prime weekday Weekend
Timeframe

Study By Metric Based Assessments LLC, 2006 Study By Metric Based Assessments LLC,
2006 25
TDS - Implication of Utilization

 A virtual cluster will be utilized more than

separate independent servers performing
the same functions
 Low intensity apps. can frequently be
consolidated.

26
TDS - Q&A

 Does a service catalogue exist which indicates

services, operating hours, and requisite assets?
 Is server utilization currently being monitored?
 Are financial interests aligned between facilities
& IT?
 Is power consumption a component of server
rationalization?
 Is virtualization technology currently in use?

27