What is cloud computing? Why is this different?
Jimmy Lin The iSchool University of Maryland Monday, March 30, 2009
Some material adapted from slides by Christophe Bisciglia, Aaron Kimball, & Sierra Michels-Slettvet, Google Distributed Computing Seminar, 2007 (licensed under Creation Commons Attribution 3.0 License) This work is licensed under a Creative Commons Attribution-Noncommercial-Share Alike 3.0 United States See http://creativecommons.org/licenses/by-nc-sa/3.0/us/ for details
What is Cloud Computing? 1.
Web-scale problems
2.
Large data centers
Source: http://www.free-pictures-photos.com/
1. “Web-Scale” Problems |
Characteristics: z z
3.
Different models of computing
4.
Highly-interactive Web applications
|
Definitely data-intensive May also be processing intensive
Examples: z z z z z z z
Crawling, indexing, searching, mining the Web Data warehouses Sensor networks “Post-genomics” life sciences research Other scientific data (physics, astronomy, etc.) Web 2.0 applications …
How much data? |
Internet archive has 2 PB of data + 20 TB/month
|
Google processes 20 PB a day (2008)
|
“all words ever spoken by human beings” ~ 5 EB
|
CERN’s LHC will generate 10-15 PB a year
|
S Sanger anticipates ti i t 6 PB off d data t iin 2009 640K ought to be enough for anybody.
Maximilien Brice, © CERN
1
Maximilien Brice, © CERN
Maximilien Brice, © CERN
There’s nothing like more data! s/inspiration/data/g;
Maximilien Brice, © CERN
What to do with more data? |
Answering factoid questions z z
(Banko and Brill, ACL 2001) (Brants et al., EMNLP 2007)
How do I make money? |
Pattern matching on the Web Works amazingly well
z z
Who shot Abraham Lincoln? → X shot Abraham Lincoln
|
Learning relations z z z
z
|
Start with seed instances Search for patterns on the Web Using patterns to find more instances
PERSON (DATE – PERSON was born in DATE
Sitting idle in existing data warehouses Overflowing out of existing data warehouses Simply being thrown away
Source of data: z z z z
Wolfgang Amadeus Mozart (1756 - 1791) Einstein was born in 1879 Birthday-of(Mozart, 1756) Birthday-of(Einstein, 1879)
Petabytes of valuable customer data…
z
OLTP User behavior logs Call-center logs Web crawls, public datasets …
|
Structured data (today) vs. unstructured data (tomorrow)
|
How can an organization derive value from all this data?
(Brill et al., TREC 2001; Lin, ACM TOIS 2007) (Agichtein and Gravano, DL 2000; Ravichandran and Hovy, ACL 2002; … )
2
2. Large Data Centers |
Web-scale problems? Throw more machines at it!
|
Centralization of resources in large data centers z z
|
Necessary ingredients: fiber, juice, and land What do Oregon, Iceland, and abandoned mines have in common?
Important Issues: z z z z z
Efficiency Redundancy Utilization Security Management overhead
Source: Harper’s (Feb, 2008)
Key Technology: Virtualization
App
App
App
Operating p g System y
App
App
App
OS
OS
OS
Hypervisor yp
Hardware
Hardware
Traditional Stack
Virtualized Stack
Maximilien Brice, © CERN
3. Different Computing Models “Why do it yourself if you can pay someone to do it for you?” |
z
z
Give me nice API and take care of the implementation Example: Google App Engine
What is the nature of future software applications? z z
Why buy machines when you can rent cycles? Examples: Amazon’s EC2, GoGrid, AppNexus
Platform as a Service (PaaS) z
|
|
Utility computing z
|
4. Web Applications
z
|
How do we deliver highly-interactive Web-based pp applications? z z
Software as a Service (SaaS) z z
Just run it for me! Example: Gmail
From the desktop to the browser SaaS == Web-based applications Examples: Google Maps, Facebook
z
AJAX (asynchronous JavaScript and XML) A hack on top of a mistake built on sand, all held together by duct tape and chewing gum? For better, or for worse…
3
What is the course about? 1.
Web-scale problems
2.
Large data centers
Web-Scale Problems? |
Don’t hold your breath: z z
3.
Different models of computing
4.
Highly-interactive Web applications
z z
|
Biocomputing Nanocomputing Quantum computing …
It all boils down to… z z
Divide-and-conquer Throwing more hardware at the problem
Simple to understand… a lifetime to master…
Divide and Conquer
Different Workers
“Work”
Partition
w1
w2
w3
“worker”
“worker”
“worker”
r1
r2
r3
“Result”
|
Different threads in the same core
|
Different cores in the same CPU
|
Different CPUs in a multi-processor system
|
Different machines in a distributed system
Combine
Flynn’s Taxonomy Instructions
Data
(Quick tour through parallel and distributed computing)
Multiple (MD)
Haven’t we been here before?
Sing gle (SD)
Single (SI)
Multiple (MI)
SISD
MISD
single-threaded process p
pipeline architecture
SIMD
MIMD
vector processing
multi-threaded processes
4
SISD
SIMD Processor
D0
D0
D0
D0
D0
D0
D0
D1
D1
D1
D1
D1
D1
D1
D2
D2
D2
D2
D2
D2
D2
D3
D3
D3
D3
D3
D3
D3
D4
D4
D4
D4
D4
D4
D4
…
…
…
…
…
…
…
Dn
Dn
Dn
Dn
Dn
Dn
Dn
Processor
D
D
D
D
D
D
D
Instructions
Instructions
MIMD Processor
D
D
D
D
D
D
D
D
D
D
Instructions Processor
D
D
D
D
Instructions
Source: MIT Open Courseware
Interface to external world
Interface to external world
Processor
Memory
Instructions
Processor Data
Data
Instructions
(Instructions and Data) Instructions
Memory y
Data
(Instructions and Data)
Processor Interface to external world
Instructions
Data
Data
Instructions
Processor
Processor
Interface to external world
Interface to external world
5
Memory
Memory
(Instructions and Data)
(Instructions and Data)
Memory
Memory
(Instructions and Data)
Instructions
Data
Data
Instructions
Processor
Processor
Interface to external world
(Instructions and Data)
Instructions Data
Processor
Interface to external world
Data
Instructions
Processor
Interface to external world
Data
Instructions
Processor
Interface to external world
Network
Interface to external world
Interface to external world
Processor
Processor Data
Data
Processor
Network
Instructions
Instructions
Data
Instructions
Memory
Memory
(Instructions and Data)
(Instructions and Data)
Choices, Choices, Choices
Interface to external world
Processor Instructions
Data
Interface to external world
Processor Data
Processor Instructions
Instructions
Data
Processor Data
Memory
Memory
(Instructions and Data)
(Instructions and Data)
Instructions
Parallelization Problems
|
Commodity vs. “exotic” hardware
|
How do we assign work units to workers?
|
Scale “up” or scale “out”
|
What if we have more work units than workers?
|
Number of machines vs. processor vs. cores
|
What if workers need to share partial results?
|
Bandwidth of memory vs. disk vs. network
|
How do we aggregate partial results?
|
Diff Different t programming i models d l
|
H How d do we kknow allll th the workers k h have fifinished? i h d?
|
What if workers die?
What is the common theme of all of these problems?
General Theme? |
Parallelization problems arise from: z z
| |
Managing Multiple Workers |
Communication between workers Access to shared resources
z z z
Thus, we need a synchronization system! This is tricky: z z
Difficult because
|
Thus, we need: z
Finding bugs is hard Solving bugs is even harder
(Often) don’t know the order in which workers run (Often) don’t know where the workers are running (Often) don’t know when workers interrupt each other
z z
Semaphores (lock, (lock unlock) Conditional variables (wait, notify, broadcast) Barriers
|
Still, lots of problems:
|
Moral of the story: be careful!
z
Deadlock, livelock, race conditions, ...
6
“Design g Patterns”
Source: Ricardo Guimarães Herrmann
master P
C
P
C
P
C
P
C
P
C
P
C
slaves
P shared queue P P
W W W W W
C
Rubber,, meet road…
C C
7
Rubber, Meet Road |
Existing tools: z z z
|
pthreads, OpenMP for multi-threaded programming MPI for clustering computing Condor, PBS, SGE, etc. for higher-level job management
The reality: z z z
Lots of one-off solutions, solutions custom code Write you own dedicated library, then program with it Burden on the programmer to explicitly manage everything
Source: Wikipedia
What’s different now?
Source: MIT Open Courseware
Questions?
8