How It Works
COM416 Cloud Computing &
Virtualization
Lecture II Cloud Computing & Virtualization 1
Today: Let’s look at some real
apps
We’ll focus on two very standard examples
Netflix movie player
Siri, Apple’s new digital companion service
How are these built?
What issues arise on the client platform?
What about in the Internet?
How is the cloud computing side structures?
Cloud Computing & Virtualization 2
Netflix App: Version 0
Plays movies on demand on a mobile
device
Server
Netflix.com
Simplest design?
• Web Services standards
• Netflix owns the data center
• Uses a fairly standard server
Cloud Computing & Virtualization 3
Version 0: Encounters issues
Hard to compete with companies that
already own massive cloud
infrastructures (Google, MSN, etc)
Web Services standards were for
downloading web pages, must adapt
them for video streams
How can we determine that the user is
legitimate?
Cloud Computing & Virtualization 4
Options for connection
Based on the Web Services standards:
Transmits web pages that contain SOAP requests: Simple Object
Access Protocol. Request could be “play movie”
The pages and responses are themselves encoded in HTML .
Requests and responses are sent using HTTP(S)
Data is moved over a TCP connection (can be insecure for HTTP,
or use the SSL security layer for HTTPS)
Dialog with the DNS maps Netflix.com to a list of IP
addresses. Client picks one
On arrival, Netflix load balancing policy routes request to
a particular server within the data center
Cloud Computing & Virtualization 5
Options for movie streaming
We could fetch segments of the movie as if they
were long web pages and render “incrementally”
This is what Netflix actually does
Convenient for users who jump around in the film
We could establish an RSS connection, in which
case a series of web pages can be transmitted by
the server, page by page
We could use a specialized streaming protocol
called AtomPub designed to improve performance
Cloud Computing & Virtualization 6
Building the App
An “App” is an application that runs in a browser
Typically, browser is told to disable its menu options and not
display a border
Gives the illusion that the App is a dedicated application, yet
in fact it can access the full power of the underlying browser
framework
Video player App?
A browser plug-in designed to work in many kinds of
browsers
Would have interactive API (“pages”) and also a player
component (code) that has plug-in “Codec” modules for the
movie format(s) Netflix supports
Cloud Computing & Virtualization 7
Picture of an App
Browser itself is a complex
machine that renders pages
but can also run code
Cookies
Cookies
Temporary
Temporary
Files
Files
Cloud Computing & Virtualization 8
A browser is a “virtual
machine”
A kind of mini-operating system
Web pages are the programs (and they can contain real
executable code)
Has various policies for which pages can access or create
which cookies (must be from same site), which files,
conditions under which user must click “ok”, etc
Intent is to protect applications from one-another and also
user from malfunctioning application
Browser security: an arms race against functionality
Most browsers have vulnerabilities and some sites use them
benignly (e.g. circumvent popup block) or maliciously
Why is web full of free porn? Hint: Many free porn sites
have code designed to seize control of your machine
Cloud Computing & Virtualization 9
Browser complexity
Browser is multithreaded and can do many things
concurrently
One page can have many frames, each with its
own security context, and each independently
active
Can execute code such as Adobe Flash, Javascript,
AJAX, Silverlight, CAJA
Some code downloads silently with web page
Other code must be installed as a “plug in” and gets
access to broader browser functionality
A plug-in “extends” the browser with new functions
Cloud Computing & Virtualization 10
Cookies
Cookies store history and other data
A file in a standard HTML format
Many possible fields, and application can add more
Browser provides the cookie for Netflix.com when
connecting to Netflix.com
To avoid huge cookies, some sites have multiple
cookies with subdomain names
Browser prevents BadGuysAreUs.com from seeing the
Netflix.com cookie as a security measure
Cloud Computing & Virtualization 11
Popups
Rendered content can generally render in
Frame that created the content
Parent frame (“entire page”)
A new frame that runs as a new tab
A new frame that runs as a new page
Frame has an associated security context (site) and can
only download or upload from that site
But since downloaded page can have new frames, and one
site can map to many places, limitation isn’t very meaningful
If frame also controls web page borders can be hard to
understand interaction as being “pages”; looks more like a
“live window” GUI
Cloud Computing & Virtualization 12
Files
When content is downloaded from the Internet,
the browser “quarantines” it by placing it in a
secure area of the file system
Intent is that only application that downloaded a
file can access it, and that files can only be created
or read from this safe part of the file system
User has a degree of control over downloaded
content but might be surprised at how much of
this there is, and what it could contain!
Cloud Computing & Virtualization 13
Localization
Our Netflix.com application wants to
stream data from:
A nearby data center
Within it, a machine with the right content
Among those, one that has light load
But Netflix won’t want to build its own
nationwide collection of data centers!
Leads to Netflix “version 1”
Cloud Computing & Virtualization 14
Netflix “outsourcing”
components
Think of Netflix in terms of main components
The API you see that runs on your client system
The routing policy used to connect you to a data center
The Netflix “home page” service in that data center
The movie you end up downloading
Netflix 1.0 breaks the solution into parts
Builds each of these aspects itself
But then pays a hosting company to run each part, and
not necessarily just one company!
Cloud Computing & Virtualization 15
Netflix Version 1
Netfli Movies
x :
Master
Home copies
Amazon.com
Cloud Computing & Virtualization 16
Features of new version
Netflix.com is actually a “pseudonym” for
Amazon.com
An IP address domain within Amazon.com
Amazon’s control over the DNS allows it to vector your
request to a nearby Amazon.com data center, then on
arrival, Amazon gateway routes request to a Netflix tier-
one cloud service component
The number of these varies elastically based on load
Netflix is experiencing
Amazon AC3 used to host the master copies of
Netflix movies
Cloud Computing & Virtualization 17
Akamai
Akamai is an example of a “content distribution
service”
A company that plays an intermediary role
Content is delivered to the service by Netflix.com
(from its Amazon.com platform)
Akamai makes copies “as needed” and distributes
them to end users who present Akamai with
appropriate URLs
Netflix.com (within Amazon.com) returns a web
page with “redirection” URLs to tell your
browser app what to fetch from Akamai
Cloud Computing & Virtualization 18
ARL (Akamized URL)
Image from Akamai’s white papers
Cloud Computing & Virtualization 19
A few options…
With Akamai, the ARL encodes information
about what the user seeks and how to find
it
Netflix.com page would be generated to
contain these Akamai ARLs using software
Akamai provides
But there have been several solutions to
this problem (we won’t get detailed due to
time limits) Cloud Computing & Virtualization 20
Netflix worry: Theft!
Digital movies cost a lot of money for Netflix
Can’t risk that people might steal them from
within Akamai or Amazon by knowing the
URLs
So Netflix uses a cryptographic encoding
scheme!
Every movie is enciphered using AES 256 coding
To decipher a movie, player must have the key
Cloud Computing & Virtualization 21
Sending key to user
We can’t just send it in plain text
Anyone on the web might see the page go by
Could use HTTPS (runs on the SSL standard)
The user’s system must log into Netflix.com.
We identify (“authenticate”) the user and verify that this
user is allowed (“authorized”) to access this movie
Web pages sent over SSL use negotiated end-to-end
security certificates (again, AES 256) hence are safe
against intrusion
So: we send the key in the web page with the ARL!
Cloud Computing & Virtualization 22
Notion of “closest” matters
We want to direct the user’s request to the
closest Amazon.com (Netflix.com) data
center
Later want to stream the movie from the
best choice of Akamai data center
But what should closest mean within the
web?
Cloud Computing & Virtualization 23
Options for identifying location
We could use GPS location
Most mobile devices now support GPS radio
They can synchronize time to within ~12ms and location
to within ~100 meters, improves over time
We can sense local wireless networks
There are companies that maintain databases
Can triangulate locations using wireless SSIDs even if
the wireless network itself is a secure one
We can ping “landmark” sites and triangulate
Cloud Computing & Virtualization 24
Pros and cons…
Laptop computers might lack GPS
information
Anyhow GPS location may not predict
network access routes: closest “within”
network could be physically far away
12 hop routes are very common
Cloud Computing & Virtualization 25
Pros and cons…
Triangulation schemes work well but not for
all parts of the world. Basically, ping each
option
Need to be reasonably “centered” relative to
landmark sites
So good for finding best Akamai data
center, but bad for predicting performance
between two computers both in Ithaca
Cloud Computing & Virtualization 26
Pros and cons…
Sensing wireless networks and IP route
“upstream”
A great way to augment GPS data, widely used by
mapping applications
Can often figure out which side of a hotel I’m in
Permits “precomputing” of best data center to use
This is a very popular approach today because
if we can precompute the data center we can
control which data center a particular user will
route to!
Cloud Computing & Virtualization 27
Identification options
Netflix identifies you You identify Netflix
IP address: not a good Netflix.com publishes the
option (mobility) “credentials” services it uses
Your name, password You go to such a service via a
automatically sent (old style secured chain of certificate
Web security, but very easy services
to compromise) It provides a public key for
Login info (annoying) Netflix
Biometrics, portable Key is public… Netflix has the
keychain “dongle” matching private key. Thus
Cookies remember login Netflix can prove that you’ve
got right site
Cloud Computing & Virtualization 28
Shazam: Flipping it around
Shazam samples music and then tells you what the
song is. It does the “reverse” of what Netflix does:
Captures fragment of music, sends it to the cloud
Precomputed search indicies are used to speed up the
search, with a fine-grained match to improve accuracy
Less common now, but coming soon: massive
parallel search in real-time within a scaled-out
cloud service
Today’s cloud is good at rapid replies but multiple
machines rarely cooperate to compute an answer in
real-time
Cloud Computing & Virtualization 29
Mobility
Suppose you grab your laptop and head
down to the local coffee shop
Reopen laptop. Can your
movie still play?
Problem: IP addresses change and in fact
the best data center choice could differ!
Cloud Computing & Virtualization 30
Network address translation
123 Main Street
You at home 207.66.18.9
192.68.1.2
Gimme! Coffee
You at Gimme! 116.71.221.17
192.68.1.3
Cloud Computing & Virtualization
31
What about Netflix.com
Netflix.com is really hosted by Amazon.com
And Amazon.com operates many (dozens)
of major data centers
Each of which has at least two IP addresses
(multihoming) to protect against ISP failures
So as you move about we see changes:
IP address / port change (Network address
translation)
Best data center to connect you to
Cloud Computing & Virtualization 32
Will your connection break?
Perhaps so
If you fetch minutes 20-26 of “Sleepless in
Seattle” you won’t care if the source changes!
In this sense you don’t have a connection
Perhaps not
We can hide the breakage of the TCP
connection and give the illusion of continuity
Can even “tunnel” TCP over a changing
connection
Cloud Computing & Virtualization 33
Resolving a web address
I need to find a web page for
www.consulting.ntchosting.c
om
Cloud Computing & Virtualization 34
Routing control
Netflix, Akamai and Amazon all have
opinions about which data center is best for
you!
Netflix: concerned about “affinity” (some server
may have best data for your account)
Amazon: concerned about load, nearness of data
center
Akamai: same issues as Amazon but also cares
about which of its servers have the movie online
Cloud Computing & Virtualization 35
Ways of doing that control
A site name like “Netflix.com” is mapped by the
DNS (Domain Name Service) to an IP address
This mapping is under control of Netflix.com itself
The “authoritative” mapping service for the domain
Each mapping has an associated lifetime
DNS uses mapping until it expires, then refreshes
DNS itself is a hierarchical structure
Thus Netflix can decide which mapping to use
Cloud Computing & Virtualization 36
In fact, Amazon plays this role
Part of the Amazon web services architecture is
to provide a service to Netflix.com that lets
Netflix advise Amazon on preferred routing
Amazon decides what routing customers “actually”
see
Akamai, in contrast, runs its own show because
Akamai isn’t pretending to be Netflix.com
Tradeoff: How frequent are these updates?
Quality of routing versus load on the DNS
Cloud Computing & Virtualization 37
Netflix Version 2... 3... 4...
Each factor of 10x brings new challenges
Netflix ends up
Owning some infrastructure (like their own
high-speed links to Latin and South America)
Renting some too
Building some of their own stuff... using some
existing stuff...
Constantly forced to rethink everything
Cloud Computing & Virtualization 38
High assurance
Could mean many things!
Netflix assurance that videos won’t be stolen and that
only authorized people can play them (more generally,
that intruders can’t steal private stuff)
The additional assurance that App won’t turn user’s
machine into a bot in some massive botnet
Technical assurance that playback won’t be interrupted
Overall goal of picking best binding for each request
Fault-tolerance: need to automatically handle things that
might happen while system is operational
Cloud Computing & Virtualization 39
What might our user desire?
She doesn’t want her movie history to be public
She isn’t at all eager for Borat to track her around the city
or to use her mobile devices to spy on her
She worries about whether her bank account is safe, and
whether she is at risk of “identity theft” when she makes
an online purchase
She depends on email yet email doesn’t always get
through reliably. And she thinks of it as private yet people
on the network could read it
She feels overwhelmed by inappropriate pop-up ads,
spam, junk web pages, and the slightest typo can take her
to a web site that “mimics” the real thing
Cloud Computing & Virtualization 40
Does she have a right to
more?
Do we have a right to privacy in the Internet?
If we’re harmed by information disclosed by a
third party, can we be repaid for the damage?
If she’s pulled over, can a policeman demand
to see her GPS data for the past half hour?
Would it be stalking if her ex-husband installed
a keyboard tracker on her PC?
Cloud Computing & Virtualization 41
Lawrence Lessig
Writes about “East Code” versus “West Code”
East code is about laws and regulations
West code is about new cyber technology ideas
He argues that we’re seeing technology get
ahead of the law and that this is having a
powerful impact on society in many
Show me
dimensions Am very much
liking Internet!
the law
Is like bedroom
and I’ll
window with
enforce it!
no curtain!
Cloud Computing & Virtualization 42
CyberWar
Richard Clarke, past national security advisor,
worries about the risk of a “Cyber War”
An adversary might wait until we depend upon all sorts
of technical systems, like a smart power grid
Then attack us by disabling those systems
Could cause physical damage: he envisages
destruction of the power grid, plane crashes, bank
failures
“We can do it. They can too.”
Also says that Chinese cyberespionage has broken
into huge range of government/military/high-tech
systems
Cloud Computing & Virtualization 43
How do they get in?
We can’t even trust the hardware!
Modern client platforms are exposed at every level
If the hardware had “extra” built-in computers that watch the ones
we run on, the network, etc, we couldn’t tell.
The network could be “virtualized” (think of the inter-frame data in
Sagan’s book/film “Contact”)
Operating systems often have vunerabilities. USB viruses exploit
the way that operating systems display disk drives.
The applications that run on them invariably have gaping holes.
Systems are misconfigured. Users pick terrible passwords, then
leave them lying around. We send sensitive data in plain text
How could we even dream of entrusting the Internet with our
most sensitive and important stuff?
Cloud Computing & Virtualization 44
Cloud: Helps, and hurts
Moving data into data centers could help
More standard, easier to manage
Could use “synthetic diversity” to repell attacks
But also can hurt
Today’s cloud systems are very weakly secured
Insider has almost unlimited potential
Operators monitor main system... but what
about backups and data “digesting”
applications?
Cloud Computing & Virtualization 45
CEO of Sun Microsystems on this issue?
"You have zero privacy anyway. Get over it."
-- Sun CEO Scott McNealy at the
launch of Jini (1/16/2009)
Cloud Computing & Virtualization 46
Google CEO Eric Schmidt
“If you have something that you don’t want anyone to know,
maybe you shouldn’t be doing it in the first place.... The reality
is that search engines do retain information... It could become
available later...”
47
Cloud Computing & Virtualization
What did Steve Jobs think?
Cloud Computing & Virtualization 48
Of course, that was before Siri
Officer: “Do you know why I
pulled you over today?”
John: “I’m sorry, sir. I don’t know how fast I
was going.”
John’s Siri: “John, your speed was 82mph, and
your top speed today was 91mph. Resume
“text Sally”?”
Cloud Computing & Virtualization 49
We’re exaggerating... a little...
Siri shouldn’t respond to “Sir” but mistake is
plausible
Siri captures voice snippets on your iPhone, GPS
data...
Ships the data to an Apple data center
Uses a mixture of cutting edge AI/NLP with a vast
database of utterances to make sense of what
you said
Remembers information to improve responses
Most mobile phones have similar capabilities!
Cloud Computing & Virtualization 50
Self-censorship
Applications like Siri force cloud computing systems
to censor themselves
They capture personal information (lots of it)
And they should use it on your behalf
But aren’t supposed to “mix” personal data (e.g.
John shouldn’t be able to query Sally’s Siri data)
This is hard because we are still far from true “AI”
that can actually understand the meaning of things
Cloud Computing & Virtualization 51
Search for your SS#
It won’t pull up any pages
But the reason isn’t that there is nothing in the
cloud that has your SS# on it
Search engines “hide” such pages to protect
us!
In fact huge amounts of information about us
flows into the cloud today. Google does have it.
So does MSN, Yahoo!, and many other
“aggregators”
Tomorrow it will only get more extreme!
Danger is that the system won’t know what to
hide
Cloud Computing & Virtualization 52
Exposes various roads we could
explore
Could think hard about meaning of
“assurance” in practice, for a world that
increasingly depends on technologies that
lack assurance properties
Could ask what the law really should
guarantee (and what about in other
countries with other laws?)
Could just ask if we can fix the technology
at least have better properties
Cloud Computing & Virtualization 53
COM 416 approach?
We don’t have time to explore every aspect
So we’ll focus on technology
Tomorrow, perhaps better laws will protect
cyberspace. But can it be done?
If we had to, could we build high assurance
cloud-hosted systems?
Stick around and find out!
Cloud Computing & Virtualization 54
