Improving Datacenter

Efficiency

Nagaraj S
Regional Manager
Systems Business
Sun Microsystems India Pvt Ltd
1
Agenda
• Trends in Today's Market
• Rethinking Processors
• Rethinking Storage
• Rethinking Servers

3
The IT Dilemma

“I want new
Business Needs services quickly”

“I want to reduce
Financial Needs my power bill”

“I'm running out of space, power

Operational Needs and cooling capacity”

4
Demand and Capacity are Colliding...
...and datacenters are right in the MIDDLE!
Watts per
• Demand Square Foot
800 • Power
• Users • Costs
• Services • Space
• Access • Heat

120
40

2003 2005 Next

Generation
Datacenter
5
Rack Density Increasing

2000 2002 2005 2008

28x2U Servers 42x1U Servers 6 BladeCenters 6 BladeCenters
2kW Heat Load 6kW Heat Load 24kW Heat Load 30kW Heat Load

Source: Emerson Network Power/Liebert

6
Not Enough Cooling for High Loads

Source: Uptime Institute

7
Not Enough Space
At the end of a dock
instead of the end of a street
• Tier1-Tier3 ECO Datacenters
at US & International Ports
• Capacity: 4000 racks and over
350 Sun Modular Datacenters
• 75MW of power, free cooling
from ocean water
• Six months time to market,
up to 40% less than
traditional build
8
The Opportunity

Increase Improve Space Cut Energy

Utilization Efficiency Costs in Half
Levels to as by 2x
High as 85%

9
Strategy: Addresses Energy Efficiency
Power savings can be addressed in many different ways
Authority to
Change Top 10 Strategies Potential Independent Energy Savings
Point Projects Entire DC
Cooling Efficiency 30%-50% 5%-30%
IT/DC Control

Power Efficiency/Distribution 20%-40% 5%-25%

Storage Efficiency 20%-75% 5%-20%
Turn on CPU Power Management 5%-40% 5%-15%
Match Infrastructure to SLAs 20%-30% 10%-15%
Retire Unused Systems 50%-100% 5%-10%
Power Efficient Systems for
New Deployments 40%-60% 10%-30%
BU Control

Technology Refresh 50%-80% 30%-50%

Optimize System Utilization:
Consolidation/ Virtualization/Workload Mgmt 50%-80% 30%-60%
More Efficient Applications 50%-95% 30%-70%

10
Strategy: Addresses Energy Efficiency
Power savings can be addressed in many different ways
Authority
to Change Top 10 Strategies Potential Independent Energy Savings
Point Projects Entire DC
Cooling Efficiency 30%-50% 5%-30%
IT/DC Control

Power Efficiency/Distribution 20%-40% 5%-25%

Storage Efficiency 20%-75% 5%-20%
Turn on CPU Power Management 5%-40% 5%-15%
Match Infrastructure to SLAs 20%-30% 10%-15%
Retire Unused Systems 50%-100% 5%-10%
Power Efficient Systems for
New Deployments 40%-60% 10%-30%
BU Control

Technology Refresh 50%-80% 30%-50%

Optimize System Utilization:
Consolidation/ Virtualization/Workload Mgmt 50%-80% 30%-60%
More Efficient Applications 50%-95% 30%-70%

11
 Why Applications Don’t Perform
Waiting for DATA – Memory can't keep up

10000
2x Every
CPU Frequency Two Years
DRAM Speeds
Relative Performance

1000

Gap
100

10
2x Every
Six Years
1
1980 1985 1990 1995 2000 2005
12
Single Threading
Up to 85% Cycles Waiting for Memory HURRY
UP AND
Single Threaded WAIT!
Performance

Typical Processor
Utilization:15–25%
Thread

C M C M C M

Time
Memory Latency Compute

13
 The Power of Chip Multi-Threading
CoolThreads !!!
UltraSPARC
Single Threaded T1 & T2
Processor Utilization:
Performance Chip Multi-threaded
Up to 85% (CMT) Performance

Thread 4 C M C M C M
Thread 3 C M C M C M
Thread 2 C M C M C M
Thread 1 C M C M C M
Time
Memory Latency Compute

14
Lowest Wattage per Thread and Core

2nd Gen. Opteron™ Xeon® 5300 POWER6 UltraSPARC T2

Threads per CPU 2 4 4 64
Watts per thread 60 30 33 1.5
Cores per CPU 2 4 2 8
Watts per core 60 30 65 12
10 GbE* N/A N/A N/A Integrated
Crypto accelerator* N/A N/A N/A Integrated
PCI-Express* N/A N/A N/A Integrated
Wattage 119 120 130 94
*Additional watts 22 22 22 0
required
Total wattage 141 142 152 94

15
Innovation Never Stops

16
Why Applications Don’t Perform
Waiting for DATA – HDDs can't keep up

• Today’s Multi-Core, Multi-Socket

application server design are
increasingly held back by slow
storage
• When requesting data, the server
spends most of it’s time waiting
for storage
• Application performance remain
sluggish regardless of the Server
CPU horsepower
• The traditional remedy of adding
more expensive DRAM may no
longer suffice as data sets double
every 2 years

17
CPU to Storage Discontinuity
The number of HDDs needed to keep up

18
New Server Memory/Storage Hierarchy

19
Cost Effective Performance
SSDs are 70X more cost effective

• Enterprise HDD • Enterprise SSD

> 180 Write IOPS > 7,000 Write IOPS
> 320 Read IOPS > 35,000 Read IOPS
> 300 GB > 32GB
> ~18W > ~3W
• $ per IOPS: 2.43 • $ per IOPS: 0.04
20
Hybrid Storage Pool Economics
Right-Size Performance & Capacity

1
the /5th
Po
1/10 wer
the th
Cos
t

21
ZFS Turbo Charges Applications
The Hybrid Storage Pool Data Management

• ZFS automatically :
> Writes new data to a very fast SSD
pool (ZIL)
> Determines data access patterns
and stores frequently accessed
data in the L2ARC
> Bundles IO into sequential lazy
writes for more efficient use of low
cost mechanical disks

22
Sun Innovation:
Driving Efficiency at Every Level

Processor & Desktop Systems & Storage Datacenter

VIRTUALIZATION

23
What Can We Offer ?
Increases Utilization Levels Up to 85%
Hard Partitions Virtual Machines OS Virtualization Virtual Tape Virtual Disk
App Identity File Web Mail Calendar Web App
Database Database Back Up DB Business
Server Server Server Server Server Server Server Archive
Apps

Mainframe / Open Systems Mainframe / Open Systems OS

Systems

Multiple OSs Single OS

Trend to Flexibility Trend to Isolation Policy-Based Data Universal Volume Manager

Management (UVM)
Very high RAS Ability to live migrate Scalable, low overhead
Highly Scalable Tiered Storage Manager
Very scalable Single OS to manage Architecture
Ability to run different (TSM)
Mature technology OS versions and types Divides system and Tape-Stacking Virtual Partition Manager
Ability to run different app administration Functionality (VPM)
OS versions De-couples OS and HW Fine grained Disaster Dynamic Provisioning
versions resource management Recovery Utility

24
Offerings from Sun
Sun Servers

Sun SPARC
Enterprise Servers
Most powerful, robust, mission-
critical virtualization servers
High performance per watt
Dynamic Domains
Solaris Containers
Improve security
CoolThreads Servers
Integrated, open source
virtualization – Logical Domains
Solaris Containers
Record-breaking performance
Breakthrough energy and space
efficiency
Save on CAPex and Opex
Available in racks/blades
x64/x86 Servers
Most powerful, scalable,
virtualized designs
Efficient Chassis Designs
Choice of hypervisor and OS
allows for investment protection
Windows
Linux
Solaris – Solaris Containers
xVM Server
VMware

25
Offerings from Sun
Sun Storage
Primary Storage
Consolidate management of
all heterogeneous storage
Greater utilization through
thin provisioning and virtual
volumes
Simplified configuration
management and
troubleshooting via Common
Array Manager (CAM)
Virtual Tape
Improved reliable backup
completions via VTL
Greater efficiencies with tape
processing via VSM
Better tape utilization and
management ease
Lower TCO, higher ROI
Tape Libraries
Capacity on demand via tape
thin provisioning
Easy creation of logical library,
drives, and media for testing
Aids in compliance
Share tape library amongst
multiple backup applications and
UNIX, Windows, AS/400,
mainframe
26
SPARC and Solaris Systems
#1 in Key Benchmarks
●
#1 in OLTP Database: TPC-C
●
#1 in Java Applications : SPECjAppServer, SPECjbb2005
●
#1 in Web/Network: SPECweb
●
#1 in Oracle BI EE
●
#1 in Oracle Hyperion
●
#1 in PeopleSoft Payroll
●
#1 in Siebel CRM
●
#1 in SAP ERP (using the Oracle Database)
Nagaraj S
nagaraj@sun.com

28