Data Center Technologies - A Case Study

(1)

(2)

(3)

Cloud Driver(s)

•

Supply & Demand Side

•

Private vs Public Clouds

•

Consolidation

•

Cloud economics will drive

(4)

(5)

THE HORSELESS CARRIAGE SYNDROME

Backward Thinking

(6)

THE HORSELESS CARRIAGE SYNDROME

"If I had asked my customers what they wanted, they would have said a faster horse."

- Henry Ford

“There will never be more than 1 million units because of the limited availability of good drivers”

- Daimler Company

Marketing Engineers Customers

“The horse is here to stay but the automobile is only a novelty, a fad.”

- Bank analyst

Analysts

…designed whip holders into the automobiles for the first 6-7 years, even though there was no horse…

(there were 8 million by 1918, over 600 million today)

(7)

MAINFRAME

CLIENT-SERVER

CLOUD

COMPUTING PARADIGM DISRUPTIONS

TECHNOLOGY ECONOMIC

(8)

MAINFRAME

• Centralized compute & storage, thin clients

CLIENT-SERVER

• PCs and servers for distributed compute, storage, etc.

CLOUD

• Large DCs, commodity HW, scale-out, devices

COMPUTING PARADIGM DISRUPTIONS

(9)

MAINFRAME

• Centralized compute & storage, thin clients

• Optimized for efficiency due to high cost

CLIENT-SERVER

• PCs and servers for distributed compute, storage, etc.

• Optimized for agility due to low cost (20-25% savings)

CLOUD

• Large DCs, commodity HW, scale-out, devices

• Order of magnitude better efficiency and agility

COMPUTING PARADIGM DISRUPTIONS

(10)

MAJORITY OF IT SPENDING (DEMAND SIDE)

JUST KEEPING THE LIGHTS ON

53% 36% 11%

Current IT Spending

New App Development

Existing App Maintenance

Infrastructure

(11)

1. SUPPLY-SIDE ECONOMIES OF SCALE

Larger datacenters have almost 50% lower TCO per server

• Server hardware costs (~45%)

• Facility & operations (~25%)

• Hardware labor costs (~15%)

• Power costs (~15%)

ANNUAL TCO/SERVER DECLINES W/SCALE

MAIN DATA CENTER COST BUCKETS

$0 $1,000 $2,000 $3,000 $4,000 $5,000 1k Server DC 100k Server DC TC O/ S e rv e r

Server Hardware Facility Hardware Operations Power

$2,361 $4,449

(12)

2. DEMAND SIDE ECONOMIES OF SCALE

Average server utilization rates are 5-10%, improve to 75% IS a 7x improvement in cost*

0% 25% 50% 75% 100% CPU U til iza ti on Time

(13)

SOURCES OF VARIABILITY (3)

(14)

Jan. 2009 Jan. 2010 turbotax.com taxcut.com hrblock.com taxact.com 5.0 2.5 0.0

RETAIL WEBSITES - DAILY HITS TAX PREPARATION WEBSITES - DAILY HITS

target.com walmart.com toysrus.com barnesandnoble.com 100 50 0 Jan. 2009 Jan. 2010 Source: Alexa ~4x normal load (Holiday shopping) ~10x normal load (Tax season) Source: Alexa

SOURCES OF VARIABILITY (3)

INDUSTRY

(15)

ONLINE PRESENTATION SITE ANIMOTO

SCALED FROM 25K TO 250K IN 3 DAYS

2,500 5,000 7,500

1 2 3 4 5 6 7 8 9 10 11 12

Months from Planning Date

Expected Upside

THIS UNCERTAIN GROWTH REQUIRES OVERPROVISIONING OF RESOURCES

Overprovisioning

Source: RightScale Source: CSG

SOURCES OF VARIABILITY (3)

(16)

Instance Client Admins Resources Instance Client Admins Resources Instance Admins Resources SINGLE-TENANT APPLICATION

3. BENEFITS OF MULTI-TENANCY

(17)

Instance Admins Resources Instance Admins Resources Instance Admins Resources MULTI-TENANT APPLICATION

3. BENEFITS OF MULTI-TENANCY

(18)

(19)

ARE PRIVATE CLOUDS A

SIGNIFICANT LONG TERM

(20)

PUBLIC vs. PRIVATE CLOUDS

• Private clouds prohibitively expensive for companies with a small install base; • Large enterprise, 1,000 servers, private feasible but at a 10x cost premium; • Public clouds bring higher scale to bear on all sources of variability;

(21)

Pu bl ic Cl oud Ec onomics

Private Cloud Preference

PUBLIC vs. PRIVATE BY SEGMENT

• Increased public cloud scale • Technology improvements • Increasing comfort • Decentralized IT • New public-only services

(22)

WHO WILL RUN THE WORLD’S

DATACENTER?

(23)

WHO WILL RUN THE WORLD’S DATA CENTER?

3 or 300?...

To what degree does economies of scale limit the number

of companies that can participate (survive) in this market?

• COGS – 20% advantage for an operator of 10DC vs single DC; • 50% reduction in G&A (legal, marketing, finance, site planning)

by operating 10 sites, resulting in another 15% savings.

• Latency requirements require 10-15 strategically located DCs

• 100ms response time is currently acceptable: 2 DCs per region;

• Bar has been rising to 50ms: 8 to 10 DCs;

• Voice translation, stock trading, etc. require 25-40ms: 30 DCs;

(24)

$0 $1,000 $2,000 $3,000 $4,000 $5,000 0 20,000 40,000 60,000 80,000 100,000 Revenue @ 8.5c/hr TCO/Server ($1,000) $0 $1,000 $2,000 $3,000 0 20,000 40,000 60,000 80,000 100,000 Profit per server

Source: CSG

INCREASING PROFIT PER SERVER INCREASING REVENUE PER SERVER,

DECLINING COST PER SERVER

Source: CSG

(25)

INDUSTRY STRUCTURE SUPPORTS ONLY FEW PUBLIC CLOUDS

Economies of scale, customer preference for global providers, and latency limitations indicate that fewer than 10 large public clouds will survive

= Efficient-Scale Data center (100,000+ servers)

10-15 Sites Put 80% of Economic Value within 25ms

Global Demand 10M servers Min Hoster Size ÷ 1-1.5M servers Max. Hosters 7-10 providers

(26)

DATA CENTER FUTURES

A BIASED VIEW

•

Power is the Oxygen for

the cloud

•

Sustainability = Low Cost

•

Whoever provides

Compute at the

(27)

(28)

DATA CENTER COSTS

$100

+

M

Quincy

10 Soccer fields, 27 MW, Hydro Power Dublin_{6 Soccer fields, 22 MW, Outside Air Economizers}

San Antonio

10 Soccer fields, 27 MW, Recycled Water

Chicago

15 Soccer fields, 60 MW, Containers, Waterside Economizers

(29)

DATA CENTER COSTS

(30)

ANNUAL CONSTRUCTION DOLLARS TRENDING UPWARDS

In 2011, Global Datacenter Construction Estimate



$50 billion

(31)

EPA PROJECTED EFFICIENCY WILL CURB DEMAND

(32)

IT’S ACTUALLY AND ECONOMICS PROBLEM, NOT A

TECHNOLOGY PROBLEM CALLED JEVONS’ PARADOX

(33)

(34)

(35)

(36)

(37)

(38)

(39)

(40)

(41)

(42)

(43)

(44)

Cloud economies of scale are stronger than commonly thought, leading to very strong first mover advantages;

o TCO of a server in large 100k DC up to 50% less than 1K DC

Expect massive integration of cloud infrastructure driven by TCO advantage;

o Through efficiency, simplification, and integration;

o Lower cost = better sustainability;

Economics of public cloud infrastructure suggest a very small number (~7) of massive, global public cloud providers prevail

o Public cloud efficient scale is around 1M – 1.5M servers;

o Network effects and experience curves will tend to further

reinforce a small number of cloud providers Think of data as a form of energy distribution