Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Vamos falar de Concorrência
Search
Plataformatec
August 31, 2012
Programming
25k
3
Share
Embed
Copy iframe code
Copy JS code
Copy link
Start on current slide
Vamos falar de Concorrência
Por José Valim, na RubyConf Brasil 2012.
Plataformatec
August 31, 2012
More Decks by Plataformatec
See All by Plataformatec
O case da Plataformatec com o Elixir - Como uma empresa brasileira criou uma linguagem que é usada no mundo inteiro @ Elixir Brasil 2019
plataformatec
5
1.2k
O case da Plataformatec com o Elixir - Como uma empresa brasileira criou uma linguagem que é usada no mundo inteiro @ QCon SP 2018
plataformatec
1
260
Elixir @ iMasters Intercon 2016
plataformatec
1
290
GenStage and Flow by @josevalim at ElixirConf
plataformatec
17
2.9k
Elixir: Programação Funcional e Pragmática @ 2º Tech Day Curitiba
plataformatec
2
340
Elixir: Programação Funcional e Pragmática @ Encontro Locaweb 2016
plataformatec
4
330
What's ahead for Elixir: v1.2 and GenRouter
plataformatec
15
2.2k
Arquiteturas Comuns de Apps Rails @ RubyConf BR 2015
plataformatec
6
410
Pirâmide de testes, escrevendo testes com qualidade @ RubyConf 2015
plataformatec
10
2.5k
Other Decks in Programming
See All in Programming
Go言語とトイモデルで学ぶTransformerの気持ち / fukuokago23-transformer
monochromegane
0
190
人間の目はかわらない、だからJPEGは30年もつ
yuzneri
12
19k
What's New in Android 2026
veronikapj
0
270
生成AIで帳票OCRが「簡単に」作れる時代になった?
kon_shou
0
1.1k
S3 を使うアプリケーションをローカル完結で動かすことに全力を注いでみた / Running S3 Apps Offline
contour_gara
0
560
ドリフトを絶対に許さない(?)CDK運用 / CDK Ops with Zero Tolerance for Drifts (?)
akihisaikeda
1
220
引き算の組織 ― アウトカムとAIに全振りするために辞めたこと ― / Organization by Subtraction
hirokiyamamoto14
PRO
0
260
komatsuna「分散システムにおけるバグ分析手法」
komatsunaqa
0
270
Gmail/Google DriveをトリガーにAIエージェントを動かそう! / Run AI agents with Gmail/Google Drive as triggers!
har1101
1
190
Pythonの実行はどこまで賢くなったのか? CPythonとPyPyから見る最適化のしくみ
curekoshimizu
0
190
言葉の格闘技のススメ~紙とペンと言葉から始める、キャリアの描き方~
progresscicada
2
170
今さら聞けない .NET CLI
htkym
0
200
Featured
See All Featured
Darren the Foodie - Storyboard
khoart
PRO
3
3.7k
Keith and Marios Guide to Fast Websites
keithpitt
413
23k
Leo the Paperboy
mayatellez
8
2.2k
Jamie Indigo - Trashchat’s Guide to Black Boxes: Technical SEO Tactics for LLMs
techseoconnect
PRO
0
630
Site-Speed That Sticks
csswizardry
13
1.5k
Put a Button on it: Removing Barriers to Going Fast.
kastner
60
4.5k
Intergalactic Javascript Robots from Outer Space
tanoku
273
27k
Product Roadmaps are Hard
iamctodd
55
12k
Leading Effective Engineering Teams in the AI Era
addyosmani
9
2.4k
Conquering PDFs: document understanding beyond plain text
inesmontani
PRO
4
3k
XXLCSS - How to scale CSS and keep your sanity
sugarenia
249
1.3M
How To Speak Unicorn (iThemes Webinar)
marktimemedia
1
540
Transcript
CONCORRÊNCIA VAMOS FALAR SOBRE
None
Core Team Member
None
Executar duas ou mais tarefas de forma simultânea
Server Server Server Server
Server
MULTI CORE ÚNICO PROCESSO
estado concorrência off off
O modelo declarativo porque matemática rocks!
•não existe “mutação” •não existe concorrência •apenas funções
factorial = lambda do |x| case x when 0 1
else x * factorial.(x-1) end end print factorial.(10) # => 3628800
Determinismo As mesmas entradas => as mesmas saídas
•não existe random() •não existe I/O em disco •não existem
efeitos colaterais •sempre o mesmo resultado
lambda do |a, b| c = expensive_function.(a) d = also_expensive.(b)
c + d end
lambda do |a, b| c = expensive_function.(a) d = also_expensive.(b)
c + d end a = 1
lambda do |1, b| c = 42 d = also_expensive.(b)
c + d end
lambda do |a, b| c = expensive_function.(a) d = also_expensive.(b)
c + d end
lambda do |a, b| d = also_expensive.(b) c = expensive_function.(a)
c + d end
Haskell Usando determinismo para performance e expressividade
l = lambda { |a,b,c| a + b + c
} l.(1, 2, 3) #=> 6 l.curry #=> #<Proc> l.curry.(1) #=> #<Proc> l.curry.(1).(2).(3) #=> 6 Currying
l = lambda { |a,b| a * b } double
= l.curry.(2) triple = l.curry.(3) Currying
mult a b = a * b double = mult
2 (mult 2 3) ((mult 2) 3) Currying haskell
mult a b = a * b double = mult
2 double 3 haskell
mult a b = a * b double = mult
2 double 3 Compilador haskell
mult a b = a * b double = mult
2 double 3 Compilador (mult 2 3) haskell
CONCORRÊNCIA VAMOS ATIVAR
estado concorrência off on
Variáveis dataflow Concorrência sem dor!
lambda do |a, b| c = expensive_function.(a) d = also_expensive.(b)
c + d end
lambda do |a, b| thread { c = expensive_function.(a) }
thread { d = also_expensive.(b) } c + d end
main
main spawn thread 1
main spawn thread 1 spawn thread 2
main spawn thread 1 spawn thread 2 unbound c
main spawn thread 1 spawn thread 2 unbound c defines
c
main spawn thread 1 spawn thread 2 unbound c defines
c unbound d
main spawn thread 1 spawn thread 2 unbound c defines
c unbound d defines d
main spawn thread 1 spawn thread 2 unbound c defines
c unbound d defines d c + d
ESTADO VAMOS ATIVAR
estado concorrência on on
None
O PROBLEMA
class Counter mattr_accessor :i self.i = 0 end thread {
Counter.i = Counter.i + 1 }
thread 1 Counter.i thread 2
thread 1 Counter.i thread 2 0
thread 1 Counter.i thread 2 0 0
thread 1 Counter.i thread 2 0 0 1
thread 1 Counter.i thread 2 0 0 1 1
thread 1 Counter.i thread 2 0 1 0 1 1
thread 1 Counter.i thread 2 0 1 0 1 1
1
thread 1 Counter.i thread 2 0 1 0 1 1
1 1
thread 1 Counter.i thread 2 0 1 0 1 1
1 1 1
thread 1 Counter.i thread 2 0 1 0 1 1
1 1 1 2
thread 1 Counter.i thread 2 0 1 0 1 1
1 1 1 2 2
thread 1 Counter.i thread 2 0 1 0 1 1
1 1 1 2 2 2
thread 1 Counter.i thread 2 0 1 0 1 1
1 1 1 2 2 2 2
•shared-memory concurrent model •locks •transactional memory •message-passing concurrent model
•shared-memory concurrent model •locks •transactional memory •message-passing concurrent model
class Counter mattr_accessor :i self.i = 0 end thread {
Counter.i = Counter.i + 1 }
class Counter mattr_accessor :i self.i = 0 end thread {
synchronize { Counter.i = Counter.i + 1 } }
thread 1 Counter.i thread 2
thread 1 Counter.i thread 2 0
thread 1 Counter.i thread 2 0 0
thread 1 Counter.i thread 2 0 0 1
thread 1 Counter.i thread 2 0 0 1 1
thread 1 Counter.i thread 2 0 1 0 1 1
thread 1 Counter.i thread 2 0 1 0 1 1
1 synchronize {
thread 1 Counter.i thread 2 0 1 0 1 1
2 1 synchronize {
thread 1 Counter.i thread 2 0 1 0 1 1
2 1 synchronize { 2 }
thread 1 Counter.i thread 2 0 1 0 1 1
2 2 1 synchronize { 2 }
thread 1 Counter.i thread 2 0 1 0 1 1
2 2 2 1 synchronize { 2 }
thread 1 Counter.i thread 2 0 1 0 1 1
2 2 3 2 1 synchronize { 2 }
thread 1 Counter.i thread 2 0 1 0 1 1
2 2 3 3 2 1 synchronize { 2 }
+ técnica mais popular + controle explícito sobre o lock
- controle explícito sobre o lock - técnica pessimista
•shared-memory concurrent model •locks •transactional memory •message-passing concurrent model
class Counter mattr_accessor :i self.i = ref { 0 }
end thread { atomic { Counter.i = Counter.i + 1 } }
thread 1 Counter.i thread 2
thread 1 Counter.i thread 2 0
thread 1 Counter.i thread 2 0 0
thread 1 Counter.i thread 2 0 0 1
thread 1 Counter.i thread 2 0 0 1 1
thread 1 Counter.i thread 2 0 1 0 1 1
thread 1 Counter.i thread 2 0 1 0 1 1
1 atomic {
thread 1 Counter.i thread 2 0 1 0 1 1
1 1 atomic {
thread 1 Counter.i thread 2 0 1 0 1 1
1 1 1 atomic {
thread 1 Counter.i thread 2 0 1 0 1 1
1 1 2 1 atomic {
thread 1 Counter.i thread 2 0 1 0 1 1
1 1 2 2 1 atomic {
thread 1 Counter.i thread 2 0 1 0 1 1
1 1 2 2 2 1 atomic {
thread 1 Counter.i thread 2 0 1 0 1 1
1 1 2 2 2 1 atomic { }
thread 1 Counter.i thread 2 0 1 0 1 1
1 1 2 2 2 1 atomic { }
thread 1 Counter.i thread 2 0 1 0 1 1
1 1 2 2 2 1 atomic { }
thread 1 Counter.i thread 2 0 1 0 1 1
1 1 2 2 2 2 1 atomic { }
thread 1 Counter.i thread 2 0 1 0 1 1
1 1 2 2 2 2 2 1 atomic { }
thread 1 Counter.i thread 2 0 1 0 1 1
1 1 2 2 2 2 2 3 1 atomic { }
thread 1 Counter.i thread 2 0 1 0 1 1
1 1 2 2 2 2 2 3 3 1 atomic { }
+ técnica otimista + não possui deadlock nem condições de
corrida
- tentativas desnecessárias - overhead com transações
•shared-memory concurrent model •locks •transactional memory •message-passing concurrent model
server = lambda do |i| receive when :increment server.(i+1) when
:check client <- i server.(i) else warn "unknown message" server.(i) end end server.(0)
thread { server <- :increment }
client 1 server client 2
client 1 server client 2 0
client 1 server client 2 0 :increment
client 1 server client 2 0 1 :increment
client 1 server client 2 0 1 :increment :increment
client 1 server client 2 0 2 1 :increment :increment
client 1 server client 2 0 2 1 :increment :increment
:increment
client 1 server client 2 0 2 1 3 :increment
:increment :increment
client 1 server client 2 0 2 1 3 :increment
:increment :increment :check
client 1 server client 2 0 2 1 3 3
:increment :increment :increment :check
client 1 server client 2 0 2 1 3 3
:increment :increment :increment :check
client 1 server client 2 0 2 1 3 3
3 :increment :increment :increment :check
“Do not communicate by sharing memory; instead, share memory by
communicating;” Effective Go
+ não precisa de sincronização + fácil de distribuir
- coordenação é difícil - modelagem não convencional
RESUMINDO
message-passing locks stm go, erlang ruby clojure dataflow oz
github.com/celluloid
@elixirlang / elixir-lang.org
É IMPORTANTE O COMPORTAMENTO DEFAULT
BALAS DE PRATA NÃO EXISTEM
REFERÊNCIAS Seven languages in seven weeks
Concepts, Techniques and Models of Computer Programming REFERÊNCIAS
REFERÊNCIAS Software Transactional Memory http://java.ociweb.com/mark/stm/article.html Persistent Data Structures http://www.infoq.com/presentations/Value-Identity-State-Rich-Hickey
http://plataformatec.com.br Estamos contratando!
None
? PERGUNTAS José Valim @josevalim