Cloud Computing Pay-as-you-go computing explained. Aake Edlund PDC, BalticGrid, BalticCloud and SweCloud

(1)

Cloud Computing

Pay-as-you-go computing explained

Aake Edlund

(2)

Acknowledgement and references

Alan Williamson,

www.aw20.co.uk

Berkeley RAD lab, berkeleyclouds.blogspot.com

Baltic Cloud team, cloud.balticgrid.eu

Rich Wolski and his Eucalyptus team, eucalyptus.cs.ucsb.edu

Dan Reed, Microsoft, www.hpcdan.org/reeds_ruminations/microsoft

451 group, www.451group.com

McKinsey ‘internal’ report (and RS blog response “McKinsey doesn’t

‘get ‘the cloud”)

Vaquero et al – collecting the 22+ cloud definitions

(3)

Cloud computing

– short version

Cloud = Virtualization + Automation



The cloud is IT infrastructure as a service.



The IT infrastructure is delivered as virtual

machines.



Automation moves those VMs around,

thereby providing the delivery mechanism for

the service.

(4)

Cloud computing

– longer version

“

Cloud Computing refers to both the

applications

delivered as services over the Internet

and the

hardware and systems software in the

datacenters that provide those services

…

The datacenter hardware and software is what

we will call a Cloud

…

Cloud computing has the following characteristics

1. The illusion of infinite computing resources..

2. The elimination of an up-front commitment by Cloud users..

3. The ability to pay for use … as needed…”

UC Berkeley RAD Labs

(5)

Spectrum of Clouds



Instruction Set VM (Amazon EC2)



Bytecode VM (Microsoft Azure)



Framework VM (Google App Engine)

EC2

Azure

AppEngine

Lower-level,

Less management

Higher-level,

More management

(6)

Two views

- of the same thing

On the right, Sam Johnston’s 6 layer Cloud Computing Stack

(7)

- Provisioning of computing resources: CPU, Memory, Processing

- Basically an “Operating System” on demand

- Usually billed on a per-hour usage model

Players in this space

Amazon EC2, Flexiscale, GoGrid, Joyent, AppNexeus

Management providers: RightScale, ElasticServer

Operational Differences

- Wildly different CPU/Memory specs

- Operating System differences

- Difficult/Impossible to move whole images between providers

Layer 1: Cloud Infrastructure

(8)

- Provisioning of data storage: Either file/object based or Database like

functionality.

- Billed on bandwidth and storage consumed

Players in the space: Amazon S3, Nirvanix, Mosso, Amazon’s

SimpleDB, Google’s BigTable, Azure Storage

Management Providers: Jungle Disk, Elephant Disk, PutPlace.com

Issues

- Different types of data storage models

- Limitations on the size of individual data units

- Different billing models makes it hard to do a straight comparison

- Access to the data generally uses non-standard query syntax

- No common API

- Performance issues

Layer 2: Cloud Storage

(9)

Provides a complete software stack - An IDE for the cloud

Takes care of: Runtimes, Load balancing, Resource provisioning

Players in the space

- Google App Engine - Python (initially, now also Java)

- Force.com (SalesForce)

- Microsoft’s Azure - .NET

- Heroku.com (RubyOnRails)

Issues

Different languages -- Most platforms are unary

Different operational philosophies

- Google App Engine for example doesn’t permit files

Lots of limitations in terms of deployment

Completely reliant on the provider for complete uptime and operation

Widely different billing models

Layer 3: Cloud Platform

(10)

- Applications that are completely ‘online’

- Operate on data that is stored in the ‘cloud’ or ‘ether’

- No client software generally required

- Billing: Ad. Revenue, Premium Services

Players in this space

- Google Apps - Gmail / Google Docs

- Apple’s MobileMe

- Microsoft’s Live - Hotmail, Live Spaces

- SalesForce.com

Issues

- Near on impossible to move between providers

- GMail to Hotmail requires major disruption

- End user focused

o

Consumer side of cloud computing

- Completely reliant on the provider for complete uptime and operation

Layer 4: Cloud Applications

(11)

- Provides services to which other applications can utilise

- Specific to vertical markets where most “Web2.0” standards live

- Usually free for non-commercial use

Players in the space

- Google/Yahoo Maps

- PayPal / Google Checkout

- Google / Yahoo WebServices

- Amazon Merchant Services

- Amazon Simple Queue Service

Issues/Comments

- Some “Web2.0” services have attempted a standardization path

- Most however are complete vendor lock-in

- Mashup applications utilize Cloud Service

Layer 5: Cloud Services

(12)

- Accessing the cloud

- Clients utilize standard access protocols

o

XML

o

JSON

o

REST / SOAP

Browsers

o

FireFox / IE / Chrome

Mobile clients

o

Google Android / Symbian / iPhone / J2ME

Desktop Apps

o

Google Gears / Adobe AIR / Microsoft Azure

Layer 6: Cloud Clients

(13)

Infrastructure

Storage

Platform

Applications

Services

Clients

Increasing

level of

complexity

and

vendor lock-in

Complexity and Vendor lock-in

(14)

Why now?

Experience with very large datacenters

Unprecedented economies of scale

Resource

Cost in

Medium DC

≈ 1000 servers

Cost in

Very Large DC

≈ 50,000 servers

Ratio

Network

$95 / Mbps / month $13 / Mbps / month

7.1x

Storage

$2.20 / GB / month

$0.40 / GB / month

5.7x

Administration ≈140 servers/admin >1000 servers/admin

7.1x

Price per KWH

Where

Possible Reasons Why

3.6¢

Idaho

Hydroelectric power; not sent long distance

10.0¢

California Electricity transmitted long distance over the

grid; limited transmission lines in Bay Area; no

coal fired electricity allowed in California.

(15)

Trends, margins

Jim Gray in 2003: economic necessity

mandates putting the data near the

application

, since the cost of wide-area networking has fallen more slowly

(and remains relatively higher) than all other IT hardware costs.

(16)

Why now?

Technology is ready



Pervasive broadband Internet



Fast x86 virtualization

Virtualization makes it all possible

Virtualization is now built-in to

microprocessors

o

Intel’s Virtualization (Intel VT)

o

Hardware based, very fast



Pay-as-you-go billing model



Standard software stack

(17)

Why now?

Business drive – cloud users

• Flexibility

• Eco-efficiency

• Credit crunch business imperatives

- CapEx to OpEx – pay-as-you-go

- Fixed cost to variable cost

(18)

Unused resources

Economics of Cloud Users

Pay by use instead of provisioning for peak

Static data center

Data center in the cloud

Demand Capacity Time Demand Capacity Time

(19)

Economics of Cloud Users

Heavy penalty for under-provisioning

Lost revenue

Lost users

Demand Capacity Time (days) 1 2 3 Demand Capacity Time (days) 1 2 3 Demand Capacity Time (days) 1 2 3

(20)

Trends: How is cloud

computing being used?

(21)

Trends: Cloud providers

(22)

* Extra per-hour charge

There are alternatives

(23)

1 month ~ 4 weeks ( 24 x 7 x 4 = 672 hours )

(24)

1 month ~ 4 weeks ( 24 x 7 x 4 = 672 hours )

CPU Cores differ

(25)

The devil is in the

details…

Your data

in the cloud

(26)

Adoption Challenges

Challenge

Opportunity

Availability

Multiple providers & DCs

Data lock-in

Standardization

Data Conﬁdentiality and

(27)

Vendor lock-in, finding

alternatives: Our shopping list



Charge on a per-usage



Ideally per-hour



Quickly and easily start-up instances



Under 5 minutes if possible



Offer base OS images that we have full control over



API to remotely control the environment



Ability to perform our image snapshots



Supported by 3rd party tools

(28)

Infrastructure

Storage

Platform

Applications

Services

Clients

Increasing level

of

complexity

and

vendor lock-in

Vendor lock-in, finding alternatives:

The

higher you go in the cloud taxonomy, the

higher the risk of lock-in

(29)

Vendor lock-in, or the

risk/benefit dance

Before you trust your business to the cloud, be sure you know how to

get out

- Do an X-ray of the vendor

- Multiple vendors (pros and cons)

- Pre negotiated contracts, with exit plans

Benefits vs risk

"There are APIs and platforms in the cloud world that create a walled

garden. You get the benefits of that garden, but you're also restricted.”

- Security customization increase the lock-in

- Only for short term? Smaller problem.

Standards – don’t hope for too much, or very soon

Putting our hope to standards – that will need a push from the customers,

the vendors like the lock-in

(30)

Compare to pre-Cloud

“If anything, [moving away from EC2] would be

easier than [exiting] an on-premise system”

"It's much harder to move from AIX to Sun than to

move from Amazon to FlexiScale”, better ask the

full picture question “Is there lock-in in the

choices I'm considering?”

(31)

Summary – alternative and

vendor lock-in

-

There are alternatives

to Amazon, and the list it growing

- Many

pricing models are confusing

and unclear, and hard to

compare

- Many pricing models

will change

(and are already)

- Important to get the big picture of what you need, today and

later on

- Vendor locking is

definitely a problem

to address.

- Before you jump into it.

But this is always true.

- Many of these ‘challenges’ are pain points waiting

for startups to solve…

(32)

Growth Challenges

Challenge

Opportunity

Data transfer bottlenecks FedEx-ing disks, Data

Backup/Archival

Performance

unpredictability

Improved VM support, flash

memory, scheduling VMs

Scalable storage

Invent scalable store

Bugs in large distributed

systems

Invent Debugger that relies

on Distributed VMs

Scaling quickly

Invent Auto-Scaler that relies

(33)

Policy and Business

Challenges

Challenge

Opportunity

Reputation Fate Sharing Offer reputation-guarding

services like those for email

Software Licensing

Pay-for-use licenses; Bulk

(34)

Startups opportunities



Startups and prototyping



One-off tasks



Research at scale

Cloud Killer Apps



Mobile and web applications



Extensions of desktop software



Matlab, Mathematica

(35)

Baltic Cloud

Bringing cloud computing to

the Baltic States and Belarus

Main motivations:

Low resource usage

Too steep learning curve

Low acceptance within industry

Complicated to adapt new applications



Typically, need a “project” to do that

(36)

Open source

“cloud in a box”

Our choice:

-

Great team! Rich Wolski, UCSB.

-

Integration with

-

Integration with

Many other open source components out there….

(37)

SweCloud

The Swedish Cloud Computing Infrastructure

Proposed pilot in October 2007 (Åke, Lennart and Fredrik H)

Started January 2009, SNIC financed, lead by Åke



Leveraging on the BC results

Roadmap



Connecting to BalticCloud



Running pilot application – this fall



Evaluation – September 2009

(38)

Key findings

Cloud computing is a

Cloud computing is a new way of using existing

new way of using existing

technology, driven by business value for the

whole chain, from

the providers to their users

and their user’s users.

At this point in time, cloud computing is best used

in startups and prototyping…

(39)

Expectations on 2009

“I’ll move from Mac to PC if this isn’t a paradigm shift”

Aake Edlund, April 21

stst

, 2009

 See more cloud computing being used by startups and in-house quick prototyping

 See more cloud computing being used in academia

 See more open source alternatives, and alternative cloud providers – also in academia (BalticCloud, SweCloud)

 Improved solutions addressing the dangers in cloud computing.

 Low expectations on standard APIs (takes time, not sure about the interest from industry, looks like for grids). I.e. industry will not help here, this is the competition epicenter.

 Hype, problems, business issues: Successful usage will prove the value. If it is a paradigm shift, it will be clear this year.

(40)

Missing slides

- Yes, there are “some” missing here

 _{Clouds and HPC}_{– Dan Reed take on this “Twins Separated at Birth: Cloud Computing, HPC}

and How Microsoft Is Trying To Change How We Think About Scale”, and others on ‘supercomputer as a service’ – a topic in its own

 _{Cloudbursting}_{– how companies outsource parts of their business to clouds, keeping the}

confident parts in-house

 Open Cloud Manifesto – “Dedicated to the belief that the cloud should be open”  OGF cloud API standardization efforts

 _{All the}_{industry buzz,}_{all the companies trying to give their customers their cloud offering}  _{Still ongoing discussions about}_definitions_…

 The 15 definitions of how to tell it is not a cloud, and why these definitions are semi-wrong….  Many, many, cloud related projects also in Europe, eg. OpenNebula.org – part of

RESERVOIR

 _{Some slides on ‘how in}_practice_{use AWS and Google AppEngine’ … (see BC course) - how} to do loadbalancing, databases in the cloud, mass storage in the clouds, and more about options to AWS , Google and Azure.

(41)

http://samj.net/2008/09/taxonomy-6-layer-cloud-computing-stack.html

m http://geekandpoke.typepad.com