2. “cloud computing is a model for enabling ubiquitous,
convenient, on-demand network access to a shared
pool of configurable computing resources (e.g.,
networks, servers, storage, applications and services)
that can be rapidly provisioned and released with
minimal management effort or service provider
interaction.” – NIST
Source: “The NIST Definition of Cloud Computing”, Special Publication 800-145.
3. cloud computing: a standardized IT capability (services,
software, or infrastructure) network-delivered in a metered,
fast, self-service way.
4.
5. “Public cloud is better from any economic point of view”
(or so the experts say)
6.
7.
8.
9.
10.
11.
12.
13.
14.
15. “The cloud has decimated sales of on-premises servers and
storage (Source: “The Public Cloud Services Market Will Grow
Rapidly To $236 Billion In 2020”, Forrester)
24. “Statistically speaking, everyone says cloud, cloud, cloud, but
only 1% of storage capacity in the world is in any public cloud”
Urs Holzle (interview with Business Insider) may 2015
30. “Public cloud allows for rapid provisioning, saving money by using only
the resources you need.”
31. “Public cloud allows for rapid provisioning, saving money by using only
the resources you need.”
32. “Software Scalability issues aside. I am not sure if Pokemon
Go would ever be possible if it not on the cloud. How could
you get instances up this fast. It had explosion of players in
very little time. There is no way you could have planned this
resources ahead of time. And it die down fairly quickly, which
means you would have lots of unused server if it were not for
cloud.”
33.
34.
35. source: “The cloud dividend: Economic impact of cloud computing”, Centre for Economics and Business Research, Dec 2010.
36. source: “The cloud dividend: Economic impact of cloud computing”, Centre for Economics and Business Research, Dec 2010
41. source: “How Fast are Semiconductor Prices Falling?” David M. Byrne, Federal Reserve Board, Stephen D. Oliner, AEI & UCLA, Daniel E. Sichel, Wellesley College and NBER, March 2015
It’s true that Moore’s law has provided a y/y improvement in price/performance rate of around
30%, but companies have a depreciation period to respect. This, and the immobilized capital
means that the maximum impact it may have is around 15% on hardware alone...
42. source: “Overall Data Center Costs.” James Hamilton, Amazon. http://perspectives.mvdirona.com/2010/09/overall-data-center-costs/
43.
44.
45. Source: Huan Liu, https://huanliu.wordpress.com/2014/02/26/amazon-ec2-grows-62-in-2-years/
Number of Amazon EC2 racks (est.)
49. “Public cloud has far greater utilization rates”
Fact: Absolutely not. Reported rates for Google (that has a scale
comparable to AWS) are not far from those reported for private clouds.
50. Source: NRDC
How VMware Virtualization Right-sizes IT Infrastructure to Reduce Power
Consumption; VMware, 2008AWS
67. Source: “Transportation Cost and Benefit Analysis II – Vehicle Costs” Victoria Transport Policy Institute, Dec 2015
Number of Amazon EC2 racks (est.)
77. Amazon EC2 Service Level Agreement:
“AWS will use commercially reasonable efforts to make Amazon EC2 and
Amazon EBS each available with a Monthly Uptime Percentage (defined
below) of at least 99.95%..”
“Unavailable” and “Unavailability” mean:
For Amazon EC2, when all of your running instances have no external
connectivity, for Amazon EBS, when all of your attached volumes perform
zero read write IO, with pending IO in the queue.
78. So we simply have to rewrite all the software in use.
99. “While IBM initially contributed software that was valued
at 40M$, external contributors to the project created
software representing a value of roughly 1.7B$ over the
examined period.”
Source: “Enabling knowledge creation through outsiders: towards a push model of open innovation” Spaeth, Stuermer, von Krogh. Int. J. of Technology Management 52(3/4). (2010)
101. ●Amazon AWS added 516 new features in 2014
●AWS changes code every 16 seconds
●Difficult for any vendor to match that rate
●Open Source does, through coopetition
●Microsoft has 30000 software engineers working on all its cloud
projects
●… which is approximately the number of overall contributors to an
OpenStack (& Linux) based open source public cloud
102. Open Source allows any company to afford the same
standardized platform approach of Public cloud providers
117. ● Open Source provides a wide spectrum of tools - use the one more
appropriate. Don’t stop to the first one or the most visible
● “There can be only one” is a nice line for an Hollywood movie, not
for your IT systems
● What is niche today can become mainstream tomorrow
● Most users don’t need the complexity of the largest platform. Some
necessary features can be added later as VMs on top
Thanks to the organizers. They didn’t know what they were doing when they invited me, especially since they didn’t know that I would bring more than 100 slides. I’m here to talk about economics; a subset of economics- IT economics, especially cloud economics.
Everyone defines cloud differently; so I will use NIST’s definition...
And simplify it. As a way go consume standardized IT services in a metered and self-service way. Usually we talk about “legacy”, private cloud, public cloud;
And you can see that the experts are very convinced about this.
They claimed that public cloud is the only alternative, private cloud is for suckers..
They claimed that public cloud is the only alternative, private cloud is for useless human beings,
Decimated. “To kill, destroy or remove a large proportion of.” Behold the decimation:
Eurostat, 2016
Forrester claims that public cloud will be a $236B market in 2020: it’s 6% of spending. In Europe, even less (4%)
In 2020 all public cloud spending will be just 5% of all IT.
In 2020 all public cloud spending will be just 5% of all IT.
One alternative is that everyone is an idiot, except the experts. Or maybe not.
There is nothing magic in cloud computing! There are still servers out there, just like ours (more or less)
true, but largely unnecessary
Behold the decimation:
Avg: 0.19
As for CAPEX/OPEX, the basic idea is that OPEX is better because you have not immobilized assets in your balance sheets, and you can claim higher detractions for tax purposes. But you can turn CAPEX into OPEX easily even without clouds: through leasing for example, or by renting hardware-as-a-service from a provider. Most colocation providers will also offer the service as well.
With this kind of workload, the public cloud costs are substantially higher than a traditional infrastructure...
Even when doing a full tco calculation, with power and personnel, it costs 2.5x (aws 3 year reserved instances)
So 15% per year will reduce costs by 10% max. But this chart does not tell everything… since it misses the largest expenditure- people.
And if we count personnel in, even less (3%)
Prices have gone down, but very little (10% in 2 years), more as a reaction to new entrants in the market than endogenous market forces
but the higher virtualization ratio is not a “feature”, but causes increased variability in workload performance. The 7% is AWS, according to “A Measurement Study of Server Utilization in Public Clouds” Huan Liu, Accenture
PUE=1!!!
Do you think that it is rare?
Actually most companies now have less than a rack
One alternative is that everyone is an idiot, except the experts. Or maybe not.
The belief that every company and authority of any size, any sector in the world has the same needs and capabilities of a silicon valley startup.
The
Guess you’re perplexed now.
Public cloud are just one end of a spectrum, exactly like other sectors, like mobility.
Wrong to assume that economics implies just cost or pricing. The bycicle is the cheapest mode, but range is limited (along with load capability and convenience - especially if it rains). Economics is the sum of all - role, needs, and cost.
cloud is renting software+execution; packaging in an easily consumable service
cloud is renting software+execution; packaging in an easily consumable service. We just have to rewrite everything to be cloud-native.
There are lots of apps…
And most of them are not even virtualized.
US Government accountability office
Infoworld’ Stewart Alsop predicted that the last mainframe would be plugged out in 1996 (in 2000 he “eat his own words”)
One developer for every 12 companies
OSS is inherently efficient by facilitating Lead Users in the design
Sometimes you need this complexity...
But it may be overkill if you just want to go shopping in the city center.
Cloud computing, like serverless and any new tech that comes along, is a tool. Experts that claim that it’s the only tool are wrong - the right tool for the right situation.
...but many other think otherwise. so I hope not to have bored you too much. Thanks.
The biggest problem is the “all I have is an hammer” problem; the idea that there is one magical single tool capable of doing anything.
Which ends up doing nothing very well.
Because very large systems tend to be not so profitable after all
As for scale, you can’t get a long term discount from a vendor higher than the profit margin for that product line (or it would basically subsidize you). This means that we can expect to have a 20% discount for example from Intel, but no more than that (outside of “spot” prices). Even if you do everything by yourself, you have to add the research and development costs as well (one of the reasons for the emergence of the OpenCompute project)
...ma solo fino a un certo punto: poi i vantaggi della legge di Moore sovrastano quelli dovuti all’accentramento nei grandi DC
But cloud computing in general is non-fungible outside of specific, standardized services like email or S3 storage. The standard is what creates the fungibility. You can’t move from AWS to Azure to an OpenStack provider with no costs - and this means that cloud is not a commodity
The experts say: the decrease of computing cost will lead to an increase in consumption! So cloud is destined to expand up to reaching 100% of IT!
But we see that this does not happened. Actually, there has been not a significant change in ICT spending growth except for the financial crisis.
No. The fallacy is the same: there is an unlimited demand for IT consumption units (use of cpu hours, storage units, etc.). If the limit is elsewhere, the paradox does not appear.
The
The
With a small discount and taking into account the advantage of reducing the immobilization of money, you end up with nearly the same NPV of colocation with much lower capital requirements.