Upload
alistair-croll
View
2.695
Download
2
Tags:
Embed Size (px)
Citation preview
State of the CloudEnterprise Cloud SummitNew York, NY
November 17, 2009
Wednesday, November 18, 2009
Good morning, and welcome to ECSSome introductory thoughts on clouds, how we got here, where we’re going.
Tagging#interop#ecs{
Wednesday, November 18, 2009
If you want to post pictures or comments, use #Interop and #ECS
A bit about Bitcurrent
Analysis and research of emerging technologies
Cloud computing, web performance, human/computer interaction, emergent communications technology
Wednesday, November 18, 2009
Vis
ibilit
y
Time
Technologytrigger
Peak of inflatedexpectations
Trough ofdisillusionment
Slope ofenlightenment
Plateau ofproductivity
http://www.gartner.com/pages/story.php.id.8795.s.8.jspWednesday, November 18, 2009
You’re probably all familiar with Gartner’s “Hype curve.” I’m sorry to say that, according to them, we’re at the apogee -- the peak of inflated expectations. Disillusionment awaits us. Then, of course, clouds will become a part of our lives.
The stages of griefThe loss of traditional IT.
Wednesday, November 18, 2009
I like to look at a slightly different curve. It’s the stages of grief, as IT loses its traditional environments. This loss comes from a number of things:- An inability to compete on cost versus the single-mindedness of cloud providers- The changing patterns of data, storage, and computation that put users everywhere and make workloads bursty- A newfound desire for agility and faster pace of change and experimentation
Vis
ibilit
y
Time
Denial
Anger
Bargaining
Depression
Acceptance
You are here
Wednesday, November 18, 2009
0
25
50
75
100
http://developer.amazonwebservices.com/connect/thread.jspa?messageID=150461http://www.google.com/insights/search/#q=%22Cloud%20computing%22&cmpt=q
2005 2006 2007 2008 2009
Nov: S
imple
Queue
Servi
ce β
Mar: S
imple
Stor
age S
ervic
e (S3
) β
Aug:
Elasti
c Com
pute
Cloud (
EC2) β
Dec:
Simple
DB an
noun
ced
Mar: E
lastic
IP, Ava
ilabil
ity zo
nes
Aug:
Elasti
c Bloc
k Stor
e
Oct: E
C2 SLA
, Wind
ows S
erve
r
Nov: C
loudfr
ont C
DN anno
unce
d
Jan:
AWSMan
agem
ent a
nnou
nced
Mar: R
eser
ved i
nstan
ces
May: S
calin
g, ela
stic L
B for E
CS
Oct: R
elatio
nal D
B Servi
ce β
Wednesday, November 18, 2009
Here’s a rough timeline of cloud computing’s growth in recent years.First: Mentions on Google InsightThen: Gartner’s Hype Curve, which they claim is a 2-5 year cycleThen: The introduction of various services from AmazonThen: Big computing companies that have inked deals with AmazonThen: A history of some of the vendors and their initial cloud products
What we agree on:
Wednesday, November 18, 2009
In the last year, we’ve reached agreement on several things.
We have a taxonomy.
Wednesday, November 18, 2009
For starters, we have a taxonomy -- which is good. Sadly, however, we abandon it at a moment’s notice.
Private Public
SaaS
PaaS PaaS
IaaS IaaS
Managedhosting
If you want
to talk
clouds, pick
one first.
Wednesday, November 18, 2009
If someone wants to have a conversation with me about clouds, they need to pick a tier, and a private or public model. Then we can compare facts.
JBoss ServerApplication instance
Intranet siteApplication code
Support ticketingCustom app, keybusiness process
Infrastructure cloud
(ECS, Joyent, Rackspace, Azure)
Agnostic PaaS(App Engine, Heroku,
Reasonablysmart, Azure)
“Flavored” PaaS(Quickbase, Bungee, Force.com, Webex
Connect)
Port to VM/AMI
Rewrite process
Rewrite code
Port code
Word processingStandard app,
no differentiation
SaaS(Google Apps, Office
Live, Basecamp, Freshbooks, Wufoo)
Copy content
Wednesday, November 18, 2009
What you want to move into the cloud will affect how you do it and what you move to.
Cloud responsibilities:Who owns what layer?
Hardware
Architecture
Operations
APIs
App logic
?IT user Cloud
This is whyclouds are new
Wednesday, November 18, 2009
This is managed hosting
Hardware Service provider owns it
Architecture User designs how things fit together
Operations User runs the machines
APIs Users talk to components directly
App logic User defines it, writes code
IT user Cloud
Wednesday, November 18, 2009
This is strategic outsourcing
Hardware User owns it
Architecture User designs how things fit together
Operations User runs VMs, not physical ones
APIs Users talk to components directly
App logic User defines it, writes code
IT user Cloud
Wednesday, November 18, 2009
This is an IaaS cloud
Hardware Service provider owns it
Architecture User chooses from predefined menu
Operations User runs VMs, not physical ones
APIs Users talk to components directly
App logic User defines it, writes code
IT user Cloud
Wednesday, November 18, 2009
This is a PaaS cloud
Hardware Service provider owns it
Architecture User chooses from predefined menu
Operations User runs VMs, not physical ones
APIs Users only talk to well-defined services
App logic User defines it, writes code
IT user Cloud
Wednesday, November 18, 2009
This is SaaS
Hardware Service provider owns it
Architecture User chooses from predefined menu
Operations User runs VMs, not physical ones
APIs Users only talk to well-defined services
App logic Service provider writes and maintains it
IT user Cloud
Wednesday, November 18, 2009
This is a private cloud
Hardware Service provider owns it
Architecture User chooses from predefined menu
Operations User runs VMs, not physical ones
APIs Users talk to components directly
Users only talk to well-defined services
App logic User defines it, writes code
Internal clients Internal IT
Wednesday, November 18, 2009
Denial: Just timesharing all over
SOA
Virtualization
Standardization
Automation
Insulates components from functionality
through consistent APIs
Reduces minimum order quantity; turns physical things into logical ones
Means users are OK with a menu of predefined
configurations
Increases the human-to-machine ratio & drives
marginal cost towards 0
Amazon S3 turns storage into a
service
Buy a slice for just an hour
LAMP, Rails, etc.
10x enterprise efficiency ratios
Wednesday, November 18, 2009
This is the ranting of luddites and server-huggersOf SOA, the insulation of components by consistent APIsOf virtualization,which - Reduces the minimum order quantity - Makes automation possible by making the physical logicalOf platform standardization
Denial: just for startups“[There are] 60,000 different customers across the various Amazon Web Services, and most of them are not the startups that are normally associated with on-demand computing.
Rather the biggest customers in both number and amount of computing resources consumed are divisions of banks, pharmaceuticals companies and other large corporations who try AWS once for a temporary project, and then get hooked.”
http://www.techcrunch.com/2008/04/21/who-are-the-biggest-users-of-amazon-web-services-its-not-startups/
Wednesday, November 18, 2009
Even as early as last year, Amazon reported that the majority of its users and its compute cycles were consumed by enterprise customers.
Denial: they’re not reliable
! !
""#$%&
""#'%&
""#(%&
""#)%&
""#*%&
""#+%&
""#,%&
""#"%&
-%%#%%&
./0123
452678923:
4;;!4<; =>?@0AB
.C0/0D
=2??2!4<; E4$!F9B5037 G22H5B
.@@E3HD3B
G22H5B!?D:B G22H5B!?D:B
.@9I%" JB@I%"
45267!6@:D/B!D?!@9B::>!H227#!
037!HB::D3H!KB::B9
L?D3H!M2957IMD7B!/23D:29D3H!?:0:D23?!K>!N0:AO=26?BCloud Encounters, Peter van Eijk, digitalinfrastructures.nl
Wednesday, November 18, 2009
We have decent evidence that they can be relied on. Peter van Eijk is presenting this data at CMG next month, but gave us an early look at some performance benchmarking he’s done on Watchmouse, a European testing platform.
Denial: they’re slow
Cloud Encounters, Peter van Eijk, digitalinfrastructures.nl
! !
"#$$%&'!'()%*!+,#)!-.!
'#!/)01#$!"2#34+,#$'
5%''%,!6%%,($7!0$48#,!9%''%,!,#3'($7:
;
<
=;
=<
>;
><
?;
?<
@;
@<
<;
>@ABA
>;;C
>DABA
>;;C
>CABA
>;;C
?;ABA
>;;C
=ACA
>;;C
?ACA
>;;C
<ACA
>;;C
DACA
>;;C
CACA
>;;C
==ACA
>;;C
=?ACA
>;;C
=<ACA
>;;C
=DACA
>;;C
!"##"$%&'()$*+'*&'((%&+
Connect times to Amazon Cloudfront from NYC
Wednesday, November 18, 2009
Peter’s data also shows that Amazon is making significant headway with infrastructure upgrades that improve performance.
Reality:Cloud operators have an unbeatable cost advantage.
Wednesday, November 18, 2009
At this point, it’s hard to argue that cloud operators will win on a cost basis alone.
How to think about costs
0
200,000
400,000
600,000
800,000
Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4
Cos
t
UpfrontFixedVariable
Wednesday, November 18, 2009
Any cost model consists of three kinds of spending: Upfront money spent to kick things off; fixed spending that doesn’t change whether you sell one or a billion units; and variable spending associated with the amount you sell.
IT costs: Upfront
Capital investment (often, “capex”)
Don’t overlook rewriting, retooling, retraining, data migration
For many enterprises, this is just the cost of periodic upgrades. They already have equipment.
Wednesday, November 18, 2009
IT costs: Fixed
Happen no matter what; a measure of leanness
May be shared with other activities, and therefore not eliminated (this is often invoked in defense of jobs)
Wednesday, November 18, 2009
Then there are the fixed IT costs that you can’t avoid. Clouds can drop these, but if you have IT running internal systems, they won’t magically evaporate when things move to the cloud. What’s more, clouds mean new tasks for IT -- things like provisioning, managing policy, and so on.
IT costs: Variable
Tied to delivery; a measure of efficiency
Needs to be less than the resulting revenue or you’ll be called a cost center
Enterprises underestimate the true costs of service delivery
Barry Lynn of 3Tera
0
125
250
375
500
Enterprise Cloud provider
Ser
vers
per
sys
adm
in
Wednesday, November 18, 2009
The variable costs are where clouds are really strong. This stuff is the costs that increase with service delivery volumes. Cloud operators can handle 500-1000 servers per person (they have to!) and completely automate everything. They also focus on cost measurement and accounting, which is a luxury for many enterprises but a necessity for clouds. Management software is an afterthought for many IT departments; but it’s a competitive advantage for cloud operators.
Clouds might seem pricey today
£0
£7,500,000
£15,000,000
£22,500,000
£30,000,000
Start up cost Year 2 Year 4 Year 6 Year 8 Year 10
Data Centre Cloud
2009 IDC analysis of running 100% of a big enterprise’s IT in-house vs on-demandUsed with permission. Copyright (c) IDC
Even with 3-yearrefresh cycles of 30%DC remains cheaper
Final score:DC: £15M
Cloud: £26MAfter year 3,cloud costsexceed DC
Wednesday, November 18, 2009
A naive look at clouds (intentionally naive, we should point out) from IDC says that clouds are expensive
But we’re deluded
£0
£12,500,000
£25,000,000
£37,500,000
£50,000,000
Start up cost Year 2 Year 4 Year 6 Year 8 Year 10
Data Centre Cloud
2009 IDC analysis of running 100% of a big enterprise’s IT in-house vs on-demandUsed with permission. Copyright (c) IDC
Cloud costs are dynamicso even if bad decisions
are made initially, capacitycan be ramped up linearly
Year 6 requires build-out for new facility +expensive refresh due
to limited spaceDC reaches spacecapacity in year 3,
50% refresh to high-end servers needed
Wednesday, November 18, 2009
Remember: Google gets 38% server utilization, with insane effort. So this is likely unattainable. What’s more, cloud providers are competing, and may (assuming interoperability of some kind) reduce their costs, too.
http://www.oncloudcomputing.com/en/2009/07/fronde-back-to-profit-by-cloud-computing/
Wednesday, November 18, 2009
Just how big are clouds? Consider that in July 2008, Microsoft revealed that it had 96,000 servers at the Quincy facility, consuming "about 11 megawatts"More than 80% dedicated to Microsoft's Live Search and the remaining for HotmailIn August, a really good discovery was posted to a blog called "istartedsomething.com": a screen shot of a software dashboard that illustrates power consumption and server count at each of Microsoft's fifteen data centers, caught in a Microsoft video posted to their web site.
Are you negotiating with cities & power companies?
“...Microsoft pays an annual utility bill just north of $13 million, which translates to just over 3.8 cents/kwh as opposed to 5.7 cents/kwh for the ELP rate...”
http://www.virtual-strategy.com/Features/Microsoft-and-Google-Cloud-Computing-Dominance-Through-Renewable-Energy/-5.html
Wednesday, November 18, 2009
Consider a San Antonio, Texas facility from Microsoft. http://ccr.sigcomm.org/online/files/p68-v39n1o-greenberg.pdfif the data center takes the full load of 44 megawatts at a 90% load factor, Microsoft pays an annual utility bill just north of $13 million, which translates to just over 3.8 cents/kwh as opposed to 5.7 cents/kwh for the ELP rate. To prove that these assumptions are in the ballpark, public documents from another SLP customer in the San Antonio area reveal that its overall utility rate is 3.7 cents per kwh.
http://www.galleries.com/minerals/elements/silicon/silicon.htm
Wednesday, November 18, 2009
First, consider silicon. Look at the cost/capacity tradeoff of computing, as described by Moore’s Law.
http://www.flickr.com/photos/monstershaq2000/2162386152/Wednesday, November 18, 2009
Then think about another form of sand – glass. Then look at the cost/capacity tradeoff of networking. Netflix pays $0.06 to send a movie over the Internet today, and will pay $0.03 next year.
http://www.flickr.com/photos/spacepleb/801902842/Wednesday, November 18, 2009
Finally, think about iron. And consider storage – which is dropping just as quickly.
The cloud trifecta
Wednesday, November 18, 2009
This trifecta of computing, bandwidth, and storage are driving costs down dramatically. Every time Google builds a data center, it can do more than the last one did.
Everything will be free.*
*Some restrictions apply.
Wednesday, November 18, 2009
Cloud computing is on a breakneck ride to zero marginal costs because of sand, iron, and glass. This means the raw materials of clouds will be free -- or too cheap to bill -- for many of us. (if you want to know more about this, see Chris Anderson’s Free)
So you won’t be building your own data centers
70% of the Global 1000 must
“Modify their data center facilities significantly” by 2012
Increase energy from 35 to 70 watts/sq. ft (sometimes up to 300 watts)
Gartner says to
Monitor energy use
Quantifying all capital and operation changes needed
Deploy virtualization and workload management tools
http://www.virtual-strategy.com/Features/Microsoft-and-Google-Cloud-Computing-Dominance-Through-Renewable-Energy.htmlWednesday, November 18, 2009
Energy is a huge issue. Even Gartner’s recommendations for saving energy will only temporarily solve the problem at hand, because energy costs will have to be cut by more than 50% in order to keep up
Wednesday, November 18, 2009
This is a GM prototype of a car that drives itself. It’s actually green technology. Know why? Because in the end, the greenest thing you can do for a car isn’t fuel: It’s making it not crash as much. If cars didn’t crash, we could get rid of most of their weight, which in turn would make them efficient. It turns out that IT is the key to efficiency.
If you’re using others’ servers, you’ll get VMs.
Wednesday, November 18, 2009
Let’s resign ourselves to the fact that we’ll get hardware from someone else (there’s a reason Intel is investing in cloud companies like Joyent, remember.) So how will that work? You’re going to get virtual machines, because that’s how the operators keep the costs low. IT and management, not cheaper machines, is the key to efficiency.
Wednesday, November 18, 2009
Clouds let IT focus on things that actually add business value. Very few companies have a competitive advantage because of their hardware infrastructure.
Wednesday, November 18, 2009
And they eliminate many of the tasks you really didn’t want to do anyway.
New kinds of attackThird-party access
Traveling across wiresShared infrastructure
The best infosec peopleMore automationHigh-end tools
Billing to catch use spikes
Wednesday, November 18, 2009
With hypervisors, other people involved, wires to cross, and so on, there are new vectors for attack. Those have to be compared to the more rigorous standardization that a cloud is likely to subject things to.
http://www.thewhir.com/web-hosting-news/102309_IT_Firms_Skeptical_About_Cloud_PEER_1_Study
Reason to avoid clouds23%
No opinion34%
Reason to move to clouds43%
Wednesday, November 18, 2009
In a study commissioned by PEER 1, users reported security as a big impediment to cloud adoption -- and a reason for doing so!"
Reality:It’s about services, not machines
Wednesday, November 18, 2009
While virtual machines were easy to understand and embrace, we’ve finally realized that it’s the services, not the machines, that matter.
SOA may matter more than virtualization
http://www.techcrunch.com/2009/04/16/mckinseys-cloud-computing-report-is-partly-cloudy/
S3
EC2
SimpleDB
CloudFront
SQS
RDB
Elastic MapReduce
Loadbalance
I used to think here...
...now I think out here
Wednesday, November 18, 2009
What started out as pay-by-the-drink storage (S3) and computational processing (EC2), now includes a simple database (SimpleDB), a content delivery network (CloudFront), and computer-to-computer messaging (SQS). Most recently, Amazon added a web-scale data processing engine with Amazon Elastic MapReduce. (It is a framework for accessing data stored in file systems and databases). It allows developers leverage Amazon’s cloud computing power by creating applications which process huge reservoirs of data (conveniently stored in Amazon S3) in parallel.Developers become systems integrators
Cloud processing is everywhere
Cloud Encounters, Peter van Eijk, digitalinfrastructures.nl ! !
"#$$%&'!'()%!'#!*!+($,-%!
.#+'%/!+('%!($&0%*+%+!
1('.!/(+'*$&%2!
"#$$%&'!'()%!'#!'.%!
3##,-%!455!6$,($%!
&-#7/!/#%+!$#'2
"-#7/+!&*$!8%!%9%0:1.%0%; <; =;; =<; >;; ><; ?;; ?<;
$->
$-@
/%
A0
/B
7B
A0
7+C$:
7+CA-
7+C+A
+,
.B
$D
!"#$%"&$#'()%*%$"#
!$++$),-"#.)(%"(-"##,-%
*55!%$,($%;E
A-*'!+('%!2$-
"-#7/!.*+!
4+(*$!
50%+%$&%
"-#7/!.*+!
670#5%*$!
50%+%$&%
F%1!G%*-*$/!
(+!0%*--:!A*0!
*1*:!A0#)!
670#5%
! !
"#$%&'(!%)*!+,#$-
!"##$%$&#'()**+*$,(-$.(//01/21345//)**+*$,(#(6$$6#"(-$.(
)**+*$,(#(6$$6#"(-$.(/01/3/789(:;(<79(<=<
)**+*$,(#(6$$6#"(-$.(/01/3/789(:;(<>;(<=<
./0!,##1$23!4#5!)*,,#6#5,-782232#%79#:!85*!
,#98%&#'!8'-!%&:*!-*2*'-*'%
0?/)''&"++@"+A /B$-),C$D
;<=7>?7@;=7@A@
!
!
!B'%6*52C!D*,(&$:
!E-*338C!F158&'*
!E3,#C!/#568G
;<=7>?7@H?7@A@
!
!
!B:3%*5-8:;C!/*%)*5,8'-3
!I8&48C!J358*,
!K581#6C!L#,8'-
;<=7>?7@?H7@A@ !M$:N8&C!J'-&8
;<=7>?7;;?7@A@ !+)&98(#C!F707B7
;@O7;H=7?=7@A@
!
!
!
!
!
!
!
!B:3%*5-8:C!/*%)*5,8'-3
!+#2*')8(*'C!.*':851
!.$N,&'C!J5*,8'-
!P5#'&'(*'C!/*%)*5,8'-3
!Q#)8''*3N$5(C!0#$%)!B45&98
!MR'9)*'C!P*5:8'G
!0%#91)#,:C!06*-*'
!S$5&9)C!06&%T*5,8'-
OO7@<;7@@7@A@ !M*,N#$5'*C!B$3%58,&8U
V;7@A7;<H7@A@
!
!
!
!
!
!
!B$91,8'-C!/*6!S*8,8'-
!I#'(!K#'(C!+)&'8
!/8(8'#C!Q828'
!08'!W58'9&39#C!F707B7
!0)8'()8&C!+)&'8
!0&'(82#5*C!0&'(82#5*
!0G-'*GC!B$3%58,&8
VA7@;?7H=7@A@
!
!B:3%*5-8:HC!/*%)*5,8'-3
!L85&3C!W58'9*
VA7@;?7AH7@A@
!
!
!
!
!+#,#('*C!P*5:8'G
!X&,,*C!W58'9*
!X#'-#'C!F'&%*-!K&'(-#:
!M#39#6C!"$33&8
!L8-$8C!J%8,G
VA7@;?7A?7@A@ !B$3%&'C!F707B7
VA7@;?7AV7@A@ !W,#5&-8C!F707B7
VA7@;?7O?7@A@ !08'%8!+,858C!F707B7
! !
"#$%&'(!%)*!+,#$-
!"##$%$&#'()**+*$,(-$.(//01/21345//)**+*$,(#(6$$6#"(-$.(
)**+*$,(#(6$$6#"(-$.(/01/3/789(:;(<79(<=<
)**+*$,(#(6$$6#"(-$.(/01/3/789(:;(<>;(<=<
./0!,##1$23!4#5!)*,,#6#5,-782232#%79#:!85*!
,#98%&#'!8'-!%&:*!-*2*'-*'%
0?/)''&"++@"+A /B$-),C$D
;<=7>?7@;=7@A@
!
!
!B'%6*52C!D*,(&$:
!E-*338C!F158&'*
!E3,#C!/#568G
;<=7>?7@H?7@A@
!
!
!B:3%*5-8:;C!/*%)*5,8'-3
!I8&48C!J358*,
!K581#6C!L#,8'-
;<=7>?7@?H7@A@ !M$:N8&C!J'-&8
;<=7>?7;;?7@A@ !+)&98(#C!F707B7
;@O7;H=7?=7@A@
!
!
!
!
!
!
!
!B:3%*5-8:C!/*%)*5,8'-3
!+#2*')8(*'C!.*':851
!.$N,&'C!J5*,8'-
!P5#'&'(*'C!/*%)*5,8'-3
!Q#)8''*3N$5(C!0#$%)!B45&98
!MR'9)*'C!P*5:8'G
!0%#91)#,:C!06*-*'
!S$5&9)C!06&%T*5,8'-
OO7@<;7@@7@A@ !M*,N#$5'*C!B$3%58,&8U
V;7@A7;<H7@A@
!
!
!
!
!
!
!B$91,8'-C!/*6!S*8,8'-
!I#'(!K#'(C!+)&'8
!/8(8'#C!Q828'
!08'!W58'9&39#C!F707B7
!0)8'()8&C!+)&'8
!0&'(82#5*C!0&'(82#5*
!0G-'*GC!B$3%58,&8
VA7@;?7H=7@A@
!
!B:3%*5-8:HC!/*%)*5,8'-3
!L85&3C!W58'9*
VA7@;?7AH7@A@
!
!
!
!
!+#,#('*C!P*5:8'G
!X&,,*C!W58'9*
!X#'-#'C!F'&%*-!K&'(-#:
!M#39#6C!"$33&8
!L8-$8C!J%8,G
VA7@;?7A?7@A@ !B$3%&'C!F707B7
VA7@;?7AV7@A@ !W,#5&-8C!F707B7
VA7@;?7O?7@A@ !08'%8!+,858C!F707B7
Wednesday, November 18, 2009
It’s all about the data
Wednesday, November 18, 2009
Data is the most important part of a cloud. MS fellow Jim Gray, in his 2003 analysis, said that compared to the cost of moving bytes around, everything else is effectively free.The economics of storage services like Flickr don’t hold up well to churn.
58
14 minutes6172001,920,0009600OC 192
2.2 hours1000Gbps
1 day100100 Mpbs
14 hours97631649,000155OC3
2 days2,01065128,00043T3
2 months2,4698001,2001.5T1
5 months360117500.6Home DSL
6 years3,0861,000400.04Home phone
Time/TB$/TBSent$/MbpsRent
$/monthSpeedMbpsContext
Source: TeraScale Sneakernet, Microsoft Research, Gray et. al
Moving data’s not easy
Wednesday, November 18, 2009
One dirty secret of cloud computing is that from a cost perspective, everything’s pretty much free compared to the price of moving bytes around. This means you can no more build an app that’s “half cloud” than you can be “half pregnant.”
Reality:The big guys are here.
Wednesday, November 18, 2009
Legitimacy, at the cost of FUD and a slow-down of experimentation because big vendors can promise.
No straight answer
0% 25% 50% 75% 100%
82%“In trial, implementation, or use of public clouds”
F5 Networks
47%“Won’t consider the cloud in
next 12 months”
ITI38%
“Unsure about adopting cloud services”
CIO.com8%
“Implementing cloud services”
29%“No interest in the cloud”
60%“Actively researching (cloud on
radar)”
Wednesday, November 18, 2009
http://www.thule-car-roof-boxes.co.uk/pictures/roof-box-with-roof-rack.jpgWednesday, November 18, 2009
http://www.flickr.com/photos/jumphigh/1565967960/
Wednesday, November 18, 2009
Jim Sivers reminded me recently of the paradox of choice. http://sivers.org/jamSheena Iyengar has been studying choice. For her research paper, “When Choice is Demotivating”,They set up a free tasting booth in a grocery store, with six different jams. 40% of the customers stopped to taste. 30% of those bought some.A week later, they set up the same booth in the same store, but this time with twenty-four different jams. 60% of the customers stopped to taste. But only 3% bought some!
http://sivers.org/jam
0
15
30
45
60
Stopped to taste Actually bought some
6 jams 24 jams
Wednesday, November 18, 2009
Both groups actually tasted an average of 1.5 jams. So the huge difference in buying can’t be blamed on the 24-jam customers being full. Lessons learned:Having many choices seems appealing (40% vs 60% stopped to taste)Having many choices makes them 10 times less likely to buy (30% vs 3% actually bought)Surgeon Atul Gawande found that 65% of people surveyed said if they were to get cancer, they’d want to choose their own treatment. Among people surveyed who really do have cancer, only 12% of patients want to choose their own treatment.
http://sivers.org/jam
35%
65%
Choose their own treatmentHave others choose
87%
13%
General population Cancer patients
Wednesday, November 18, 2009
Surgeon Atul Gawande found that 65% of people surveyed said if they were to get cancer, they’d want to choose their own treatment. Among people surveyed who really do have cancer, only 12% of patients want to choose their own treatment.
Composed designs replace component architectures
High performance
Compliant
“Embarrassingly distributed”
Bursty/seasonal
Resilient & highly available
Scalable but eventually consistent
Wednesday, November 18, 2009
“Private” and “hybrid” concepts emerge
Private cloud: On-premise infrastructure with cloud-like properties
Hybrid cloud: Policy-driven combination of on-premise and on-demand components
Virtual private cloud: On-premise privacy on someone else’s machines
Wednesday, November 18, 2009
Always on premise
Private
Compliance-enforced
Need to track and audit
Legislative
Data near local computation
Can be done anywhere
Testing
Training
Prototyping
Batch processing
Seasonal load
Always in cloud
Partner access
Proximity to cloud services (storage,
CDN, etc.)
Massively grid/parallel (genomic,
modelling)Lo
ad/p
ricin
g en
gine
Polic
y en
gine
Virtual machine(infrastructure cloud)
Compute task(service cloud)
Wednesday, November 18, 2009
Going forward, we’ll see hybrid on-premise/on demand hybrid clouds that can intelligently move processing tasks between private an public infrastructure according to performance requirements, pricing policies, and security restrictions.
Better economics
Pay-per-use pricing
No capital investment
No long term contracts
Ideal for spiky applications
Optimized for Web 2.0 apps
Scales easily*
James Staten, Forrester* Easy scaling may not be included
Developer empowerment
Self-service portal
Infrastructure managed by cloud provider
Developer-ready framework
For all levels of developers
Cheap test and dev
One button deploy
Wednesday, November 18, 2009
Clouds promise a lot: James Staten of Forrester loves clouds, not only for the economies of scale they offer, but also for the way in which they empower developers to build and experiment by speeding up the IT cycle time.
Minimize maintenance costs Cooling Electricity Servers maintenance, backups, etc.
Elasticity and scalability Massive scale leads to true economies of scale Eliminate need to build for infrequent peaks Make capacity available on demand
Ops cost reduction Flat data sets Streamlined data management Data availability - enabling information generation
DR cost reduction
Do new things Perform mission tasks that were not able to achieve otherwise New tools like Hadoop and Map Reduce allows for amazing processing
Speed up the organization Faster, cheaper innovation Transform how gov does business Prototyping enablement
Publish databases Reduce start-up
Work differently Realtime collaboration Ubiquitous access to unlimited amount of computing power Ubiquitous access to unlimited amount of storage
Enterprise tech Disruptive tech
Rod Fontecilla, Booz, Allen, HamiltonWednesday, November 18, 2009
•60 seconds per page
•200 machine instances
•1,407 hours of virtual machine time
•Searchable database available 26 hours later
•$144.62 total cost
Desktop EC2
Pages 17,481 17,481
Minutes/page 1 1
# of machines 1 200
Total minutes 17,481
Total hours 291.4 26.0
Total days 12.1 1.1
Wednesday, November 18, 2009
One of the most interesting uses of cloud computing is time dilation. Okay, not really, but close: The Washington Post, needed to get all 17,481 pages of Hillary Clinton’s White House schedule scanned and searchable quickly. Using 200 machines, the Post was able to get the data to reporters in only 26 hours. In fact, the experiment is even more compelling: Desktop OCR took about 30 minutes per page to properly scan, read, resize, and format each page – which means that it would have taken nearly a year, and cost $123 in power, to do the work on a single machine.
Two kinds of data centerReally big data centers for really big problems
Tens of thousands or more servers
Tens of Mega-Watts of power at peak
Aimed at massive data analysis applications (search indexes, social media, genomics)
Variety of workloads
Huge amounts of fast RAM
Massive numbers of CPU cycles
High-volume disk I/O bandwidth
Requires lots of communication between servers, so network propagation affects computation speed
Micro data centers for “embarrassingly distributed” applications
Thousands of servers
100s of kilowatts.
Aimed at highly interactive apps (Interactive, office productivity apps)
Placed close to populations to minimize network transit impact
The Cost of a Cloud: Research Problems in Data Center NetworksAlbert Greenberg, James Hamilton, David A. Maltz, Parveen Patel Microsoft Research, Redmond, WA, USA
Wednesday, November 18, 2009
Sharing > Protection
Drew Bartkiewicz, The Hartford, quoted in Unseen Liability
Wednesday, November 18, 2009
According to Drew “Bartievitz” of the Hartford, there’s a shift in the value of information assets underway.
Clouds as peripherals
Clouds as IT strategy
Wednesday, November 18, 2009
Most of the enterprises I’ve spoken with use clouds as peripherals. In the same way we used to plug peripherals into our computers, enterprises plug clouds into their IT. They might have it for backup, or messaging, or content delivery, or for a specific business process. But to really harness the power of cloud computing, enterprises need to embrace it as more than just a bunch of things to plug into the organization. It needs to become part of their strategy.
Support
Onboarding
Contracts
Policies
UI
API
Language
Protocol
ComputingStorageDelivery
Wednesday, November 18, 2009
You can target a vertical. There are always ways to specialize within a specific industry. This isn’t about the computing -- as we’ve seen, this is a commodity. But you can <click> focus on a specific language or protocol, <click> UI or API, <click>, set of contracts and policies, <click>, or even support and onboarding. Every industry or target customer has specific needs. Maybe it’s the AMQP protocol, or HTML 5 optimization, or JavaScript code, or long contract terms, or high-touch support for small businesses.
Delay (in seconds)Traffic (requests/sec)
Capactity (# of machines)=
Wednesday, November 18, 2009
There’s a basic equation in computing. Performance equals traffic divided by capacity. Put another way, more users and something gets slower. More machines and something gets faster.
Wednesday, November 18, 2009
This is an example of that relationship. As usage grows, performance gets worse.
Delay (in seconds)Traffic (requests/sec)
Capactity (# of machines)=
∞
Wednesday, November 18, 2009
But when if the capacity is infinite?
Wednesday, November 18, 2009
Then you set user experience (“under 1 second”) and the elastic platform adds capacity as needed. The only problem? The bill at the end of the month!
0
25
50
75
100
http://developer.amazonwebservices.com/connect/thread.jspa?messageID=150461http://www.google.com/insights/search/#q=%22Cloud%20computing%22&cmpt=q
2008 2009 2010 2011 2012What is
the cloud?Why
should I use it?
How do I use it?
What new things are possible?
What must I still run in-house?
Taxonomies & layers
ROI, TCO, business
cases
Designs & best
practices
Business strategy
Policy & standards
Wednesday, November 18, 2009
Here are my predictions for the next few years, and what you’ll see at conferences, in the press, and in the boardroom.
Different Clouds for Different FolksIan Knox (Skytap), Lew Moorman (Rackspace), Sesh Murthy (IBM), Scott Ryan (Asankya)
The Risks of On-Demand ComputingAnthony Arnott (Trend Micro), Drew Bartewicz (The Hartford), Marc Lindsey (Levine, Blaszak, Block & Boothby LLP)
What's Working, What's Not: A Report from Cloud Adopters
Colin Hostert (Grooveshark), Geir Magnusson (Gilt), Dominic Preuss (FiLife), Vince Stephens (Taser)
Cloud Interoperability: Do We Need It? What Would it Look Like?
Chris Brown (Opscode), Jason Hoffman (Joyent), John Willis (Zabovo)
Cloud Computing RoadmapsKen Comee (Cast Iron Systems), Morris Panner (OpenAir), Randy Bias (Cloudscaling)
Wednesday, November 18, 2009