CLOUD COMPUTING

INFRASTRUCTURE
TAKE A SEAT & PREPARE TO FLY
1 Anh M. Nguyen
CS525, UIUC, Spring 2009
GOALS
 Define
 Cloud:public cloud, private cloud
 Cloud Computing
 Why cloud computing?
 Obstacles & opportunities

 Current state of cloud computing

 Amazon EC2
 Google AppEngine

2
WHAT IS CLOUD COMPUTING?
 I don’t understand what we would do differently
in the light of Cloud Computing other than change
the wordings of some of our ads
Larry Ellision, Oracle’s CEO

 I have not heard two people say the same thing

about it [cloud]. There are multiple definitions out
there of “the cloud”
Andy Isherwood, HP’s Vice President of European Software Sales

 It’s stupidity. It’s worse than stupidity: it’s a marketing

hype campaign.
Richard Stallman, Free Software Foundation founder

3
SOFTWARE AS A SERVICE (SAAS)
 Application is used as an on demand service. Often
provided via the Internet
 Think on-demand TV programs
 Example: Google App (online office)

 Benefits to users
 Reduce expenses: multiple computers, multiple users
 Ease of usage: easy installation, access everywhere
 Benefits to providers
 Easierto maintain
 Control usage (no illegal copies)
4
UTILITY COMPUTING (UC)
 Computing resources (cpu hour, memory, network) and
platform to run software are provided as on demand
service
 Think electricity service
 The same evolution happened
 Hardware as a service (HaaS), Infrastructure as a service
(IaaS), Platform as a Service (PaaS)
 Examples of UC providers: Amazon EC2, Google
AppEngine …
 Who will use UC? Is UC the end of high-end PC?
 Peoplewho otherwise has to build their own data center:
SaaS providers, analytics & batch processing
5
UTILITY COMPUTING - BENEFIT TO
USERS
 Mitigate the risks of over-provisioning and under-
provisioning
 No up-front cost, invest on other aspects
(marketing, technology…)
 Less maintenance & operational cost

 Save time, time = money

 In summary: Reduce cost

6
UTILITY COMPUTING – MITIGATE RISKS
Capacity
 Real world utilization 5%-20%
 Animoto demand surge:
from 50 servers to 3500
servers in 3 days
 Black Friday sales
Demand
Over-provisioning t

Capacity Capacity

Demand Demand
t 1 2 3 t 7
On demand, scalable
Under-provisioning
UTILITY COMPUTING – BENEFIT TO
PROVIDERS
 Make money
 Economies of scale
Resource Cost for medium scale Cost for large scale Ratio
Network $95 / Mbps / month $13 / Mbps / month ~7x
Storage $2.20 / GB / month $0.40 / GB / month ~6x
Administration ≈140 servers/admin >1000 servers/admin ~7x

 Timediversity: different peeks for different services

 Geographical diversity: choice of best location
 Electric price in Idaho = 1/5 in Hawaii
 Existing infrastructure & expertise
 Google, Amazon: utilize off-peak capacity
8
UTILITY COMPUTING – AMAZON EC2
 Elastic Compute Cloud
 Rent virtual machine instances to run your software.
Monitor and increase / decrease the number of VMs
as demand changes
 How to use:
 Create an Amazon Machine Image (AMI): applications,
libraries, data and associated settings
 Upload AMI to Amazon S3 (simple storage service)
 Use Amazon EC2 web service to configure security and
network access
 Choose OS, start AMI instances
 Monitor & control via web interface or APIs
9
AMAZON EC2
 Characteristics:
 Elastic: increase or decrease capacity within minutes
 Monitor and control via EC2 APIs
 Completely controlled: root access to each instances
 Flexible: choose your OS, software packages…
 Redhat, Ubuntu, openSuse, Windows Sever 2003,…
 Small, large, extra large instances
 Reliable: Amazon datacenters, high availability and redundancies
 Secure: web interface to configure firewall settings
 Cost:
 CPU: small instance, $0.10 per hour for Linux, $0.125 per hour for
Windows (1.0-1.2 GHz 2007 Opteron or 2007 Xeon processor)
 Bandwidth: in $0.10, out $0.17 per GB
 Storage: $0.10 per GB-month, $0.10 per 1 million I/O requests

10
UTILITY COMPUTING - GOOGLE
APPENGINE
 Write your web program in Python and submit to
Google. It will take care of the rest
 How to use
 Download AppEngine SDK
 Develop your program locally
 A set of python programs, input = requested url, output =
return message
 Debug locally

 Registerfor an application id
 Submit your application to Google

11
GOOGLE APPENGINE – HELLO WORLD
Cre

ating a Simple Request Handler

Creat
e a file helloworld.py:
print 'Content-Type: text/plain'
print ''
print 'Hello, world!‘
Ma
p url to handler
Edit
configuration file app.yaml
application: helloworld
version: 1
handlers:
- url: /.*
script: helloworld.py
Dat
a storage:
 Distributed file system
 Store using AppEngine API, retrieve using GQL
De
bug: http://localhost:8080/ 12
GOOGLE APPENGINE
R

egister for an application ID

 http://appengine.google.com
 Verification code sent to your mobile

U

ploading the Application

 appcfg.py update helloworld/
 Enter your Google username and password at the prompts
 http://application-id.appspot.com

M
anage using Administration Console
 Set up domain name
 Invite other people to be developers
 View error logs, traffic logs
 Switch between different versions 13
GOOGLE APPENGINE
 Characteristics
 Easy to start, little administration
 Scale automatically
 Reliable
 Integrate with Google user service: get user nickname, request
login…
 Cost:
 Can set daily quota
 CPU hour: 1.2 GHz Intel x86 processor
 Free quotas going to be reduced soon

Resource Unit Unit cost Free (daily)

Outgoing Bandwidth gigabytes $0.12 10GB
Incoming Bandwidth gigabytes $0.10 10GB
CPU Time CPU hours $0.10 46 hours
14
Stored Data gigabytes per month $0.15 1GB (all)
SPECTRUM OF ABSTRACTIONS
 Different levels of abstraction
 Instruction
Set VM: Amazon EC2
 Framework VM: Google AppEngine
 Similar to languages
 Higher level abstractions can be built on top of lower
ones
Lower-level, Higher-level,
More flexibility, Less flexibility,
More management Less management
Not scalable by default Automatically scalable

15
EC2 Azure AppEngine
Force.com
 DETAILED COMPARISON
Amazon Google AppEngine
Computation •x86 Instruction Set Architecture •Predefined 3-tiers Web app
model •Not scalable by default. Can structure
use 3rd party service such as •Fixed language: Python
RightScale •Automatic scaling up and
down
Storage model•Scaling varies from none (EBS) •Fixed API: BigTable
to fully automatic (SimpleDB, S3) •Automatic scaling

Networking •Define network access policies •Fixed topology to for 3-tier

model •Choose availability zones, Web app structure
independent network failure •Automatic scaling
•Elastic IP addresses, persistently
routable name
•Automatic scaling

16
WHAT IS A CLOUD?
 Software and hardware to operate datacenters
 Public cloud: cloud used to provide utility computing
 Amazon EC2: Amazon datacenters, Xen, EC2 APIs and
administrative interface
 Google AppEngine: Google data center, GFS,
AppEngine APIs, administrative interface…
 Batch processing softwares: MapReduce, Hadoop, Pig,
Dryad
 Private cloud: datacenters, not available for rental

 How about the academic clouds?

17
 Protected clouds
 Cloud Computing
A combination of existing concepts

SaaS Users PaaS Users

SaaS Utility Computing

SaaS Providers PaaS Providers

/ PaaS Users

18
 CLOUD COMPUTING
Cloud Computing = SaaS + PaaS (utility computing)

Electricity
Cloud TV Video On Demand
(Cloud Computing) (SaaS) On Demand
(PaaS)
19
WHAT IS NEW IN CLOUD COMPUTING
 The illusion of infinite computing resources
 The elimination of an up-front commitment by users

 The ability to use and pay on demand

 Cloud Computing vs P2P?

 Both take advantage of remote resources
 P2P: does not use clouds (datacenters), peers do not
get paid, lower reliability
 Cloud Computing vs Grid Computing?
 Both use clouds
 Grid Computing requires commitment, share based on 20
common interests. Not public cloud
CLOUD KILLER APPS
 Mobile and web applications
 Mobile devices: low memory & computation power
 Extensions of desktop software
 Matlab, Mathematica
 Batch processing / MapReduce
 Peter Harkins at The Washington Post: 200 EC2 instances
(1,407 server hours), convert 17,481 pages of Hillary Clinton’s
travel documents within 9 hours
 The New York Times used 100 Amazon EC2 instances +
Hadoop application to recognize 4TB of raw TIFF image into
1.1 million PDFs in 24 hours ($240)

21
SHOULD I MOVE INTO A CLOUD
 Does it really save money?

 Costcloud > Costdatacenter , balance by Utilization

 UserHourscloud > UserHoursdatacenter (under-provisioning)
 Other factors
 Re-implement programs
 Move data into cloud
 What else?
 Example:
 Upload rate 20Mbits / s. 500GB takes 55 hours
 If can process locally in less than 55 hours  moving into a cloud would not
save time

22
ADOPTION CHALLENGES
Challenge Opportunity
Availability Multiple providers
Data lock-in Standardization
Data Confidentiality and Encryption, VLANs, Firewalls
Auditability

 Coghead, a cloud vendor closed its business a week ago

 Customers need to rewrite their applications
 Online storage service The Linkup closed July 10, 2008
 20,000 paying subscribers lost their data

23

23
 ADOPTION CHALLENGES

24

Cloud Control, InformationWeek Reports, 2009

 GROWTH CHALLENGES

Challenge Opportunity

Data transfer bottlenecks FedEx-ing disks, reuse data multiple

times
Performance unpredictability Improved VM support, flash memory
Scalable storage Invent scalable storage
Bugs in large distributed Invent Debugger using Distributed VMs
systems
Scaling quickly Invent Auto-Scaler

25
GROWTH CHALLENGES
 Data transfer bottle neck
 WAN cost reduces slowest:
2003  2008: WAN 2.7x, CPU 16x, storage 10x
 Fastest way to transfer large data: send the disks

 Performance unpredictability
 Large variation in I/O operations
 Inefficiency in I/O virtualization

26
POLICY AND BUSINESS CHALLENGE
Challenge Opportunity

Reputation Fate Sharing Offer reputation-guarding services like

those for email
Software Licensing Pay-for-use licenses; Bulk use sales

 Reputation: Many blacklists use IP addresses and

IP ranges
 Software licensing:
 Open source software readily applicable
 Windows, IBM softwares offered per hour for EC2

27
THE FUTURE?
 Application software:
 Cloud & client parts, disconnection tolerance
 Infrastructure software:
 Resource accounting, VM awareness
 Hardware systems:
 Containers, energy proportionality

28
DISCUSSION
 Is their definition correct?
 What applications of cloud computing in your
research area that you can think of?
 Which service would you choose, EC2 or
GoogApp?
 Can you predict the future of cloud computing?

29
REFERENCES
 Above the Clouds: A Berkeley View of Cloud Computing, Michael
Armbrust et al, Feb 2009 (white paper and presentation)
 Google AppEngine: http://code.google.com/appengine/
 Amazon EC2: http://aws.amazon.com/ec2/
 Lessons From The Demise Of A Cloud Startup, John Foley, Feb 2009
 Cloud Control, InformationWeek Reports, 2009

30
ARE YOU READY FOR A RIDE?

31
BACKUP SLIDES

32
RIGHTSCALE
 $2500 initial fee
 $500 monthly

33