Add concurrency-article

_posts/2017-08-19-concurrency.md

@@ -0,0 +1,369 @@
+---
+ layout: post
+ title: "Concurrency-Parallelism-and beyond"
+ date: 2017-08-19
+ author: Bhavishya
+ category: Concurrency
+ tags:
+ - Concurrency
+ - Parallelism
+ - Python
+---
+
+# Concurrency…Parallelism…and Eventual world Domination;) …
+
+NOTE: “Bugs(typos?) are always there, it’s for you to find”…Same goes for this
+article.
+
+#### So what’s in-store?
+
+What I hope with this article is,“Ideally” it should be a primer for people
+starting from almost zero knowledge(or more quantitatively no practical coding
+experience implementing concurrency) to get a sufficient feel for what it is and
+from there on they might explore more at their own discretion…
+
+The structure would be to give away the theory first followed by some code-
+walks seeing concurrency live in action across different scenarios and
+paradigms…
+
+So let’s dive in…
+
+#### **CONCURRENCY VS PARALLELISM ?**
+
+> *“Concurrency is about dealing with lots of things at once. Parallelism is about
+> doing lots of things at once.” — Rob Pike*
+
+Remember “**Concurrency:Sequential :: Parallel:serial”.**
+
+So Finally once again straight from Wikipedia:
+
+> “It is possible to have parallelism without concurrency (such as bit-level
+> parallelism), and concurrency without parallelism (such as multitasking by
+time-sharing on a single-core CPU).”
+
+Concurrency and Parallelism are both independent phenomena…So a quick QUIZ:
+
+Imagine we go down 20 yrs back into history…the MULTI-CORE magic hasn’t happened
+yet…and we all have single core computers…what does that mean?
+
+It means no “parallel computing”. But still, the CPU(I’d say kernel maybe) does
+processes concurrently(Oh yes, remember multitasking).
+
+#### HISTORY?
+
+So it all began with **E.W. Dijkstra[1] (remember Shortest Path algorithm)** in
+The 1960s and then there was **Tony Hoare**(yes the guy who came up with
+“QUICKSORT”) who invented
+[CSP](https://www.wikiwand.com/en/Communicating_sequential_processes) and
+then came **Erlang(the language used by WHATSAPP’s backend**) and much later(in
+last decade)** Elixir and Clojure(**[DISCORD](https://discordapp.com/)** if you
+know is written in Elixir)**…But the true reason of the recent buzz(and somewhat
+inspiration for this post too) is [golang](https://golang.org/){This is a
+relatively new language straight from google which has brought CONCURRENCY to
+forefront again by providing some *cool constructs that make doing concurrency
+easier*…You could well do it in any other language but it would difficult(as in
+“*cry yourself* *to sleep” *difficult )}.
+
+SIDE-NOTE: For people who are aware of Functional
+Programming(Erlang, Elixir, Clojure all are functional) might have heard that
+**concurrency is easier to do in functional languages**…that is because of
+**immutability of objects**( the …fundamental feature of FP)…And one basic problem
+with concurrency is of simultaneous changing **of an object’s state(**[race
+conditions](https://stackoverflow.com/questions/34510/what-is-a-race-condition)**)**…So
+simply put "no(less) change -> easier concurrency".(This is an oversimplified
+explanation.)
+
+Here’s a fun comparison, that displays the ease of doing concurrency in different
+languages…
+
+![Plot](https://github.com/bhavishyagopesh/pclub.in/blob/concurrency-article/assets/image/concurrency-article.png)
+
+> Hey don’t get scared of the last paragraph it’s just to convince you that
+> CONCURRENCY ITSELF IS A CONCEPT AND COULD BE DONE IN MANY LANGUAGES WITH VARYING
+EASE(OR PAIN)…it’s perfectly fine if you just remember the last line…
+
+#### But What’s with all this jargon?
+
+As you shall see(0r may have already heard) there is a heck lot of nomenclature
+associated with these two concepts…*Threads, Processes,
+Gevents,greenlets,Actors,Hyperthreading,Multiprocessing,Go-Routines,Channels,AsyncIO,EventLoop…*
+
+So what are all these[2] ? …**this isn’t the motive of this article** primarily
+but we would try to understand a few of these(under specific conditions) on our
+way…
+
+**TOWARDS REAL STUFF…**
+
+Enough of talking, let’s see CONCURRENCY in action.
+
+And I choose Python as my language of choice.Yeah it’s a kinda controversial but
+I would justify this by saying that this in favor of the larger audience.Python
+is *not* a great way to do concurrency (observe its position in the
+plot)…actually, it sucks at concurrency([3])…but still…
+
+And another **warning** to make the code example a bit interesting, I’ll use the
+socket module(you might wanna see
+[this](https://pymotw.com/3/socket/addressing.html)).So things would tend to
+*get a bit dirty from here on…so hang tight and don’t be shy of googling stuff*.
+
+We’ll use Fibonacci function for demo,
+
+```python
+# Utility function for demonstration
+
+def fib(n):
+    if n <= 2:
+        return 1
+    else:
+        return fib(n-1) + fib(n-2)
+
+```
+
+And here’s the server code, it listens on localhost and sends back the fib(n)
+value:
+```python
+# server code
+
+from fib import fib
+from socket import *
+
+def fib_server(address):
+    # This sets up the socket(you can think that of as pipe over which communication happens) object
+    sock = socket(AF_INET, SOCK_STREAM)
+    sock.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
+    # It binds to the address and listens for any connections
+    sock.bind(address)
+    sock.listen(5)
+
+
+    while True:
+        client, addr = sock.accept()
+        print("Connection", addr)
+        fib_handler(client)
+
+# This function handles the connection and sends back the fib() value
+def fib_handler(client):
+    while True:
+        req = client.recv(100)
+        if not req:
+            break
+        n = int(req)
+        result = fib(n)
+        resp = str(result).encode('ascii') + b'\n'
+        client.send(resp)
+    print("Closed")
+
+# This starts a server that listens to localhost on port 25000
+fib_server(('', 25000))
+```
+
+Now start the server,
+
+    $ python server.py
+
+And in another terminal window,
+
+    $ nc localhost 25000
+      1
+      1
+      3
+      2
+      35
+      9227465
+
+Now this is all happening sequentially, So as to later compare our results we’ll
+benchmark this,
+
+```python
+# Time of a long running request
+
+from socket import *
+import time
+
+sock = socket(AF_INET, SOCK_STREAM)
+sock.connect(('localhost', 25000))
+
+while True:
+    start = time.time()
+    sock.send(b'30')
+    resp = sock.recv(100)
+    end = time.time()
+
+print(end - start)
+```
+
+    # here are the numbers:
+
+    0.2433168888092041
+    0.22079825401306152
+    0.22012019157409668
+    0.2199845314025879
+
+Now we’ll introduce threads,
+
+A thread is:
+
+* An **independent task running** inside a program
+
+* Shares resources with the main program(memory, files, network connections, etc.)
+* Has its own independent flow of execution…(stack, current instruction, etc.)
+
+By using python’s threading module you can create **hundreds** of threads and
+can give them **(target)function which they’ll execute independently of main
+program**(THIS IS partially false…becoz GIL prevents simultaneous execution — it
+here actually is preventing same shared memory to be accessed by two threads and
+cause problems like changing a variable’s value simultaneously, so GIL in that
+way is good)
+
+Here’s the threaded-server,
+
+```python
+
+# threaded-server code
+
+from fib import fib
+from socket import *
+from threading import Thread
+
+def fib_server(address):
+    sock = socket(AF_INET, SOCK_STREAM)
+    sock.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
+    sock.bind(address)
+    sock.listen(5)
+
+    while True:
+        client, addr = sock.accept()
+        print("Connection", addr)
+        Thread(target=fib_handler, args=(client,), daemon=True).start()
+
+def fib_handler(client):
+    while True:
+        req = client.recv(100)
+        if not req:
+            break
+        n = int(req)
+        result = fib(n)
+        resp = str(result).encode('ascii') + b'\n'
+        client.send(resp)
+    print("Closed")
+
+# This starts a server that listens to localhost on port 25000
+fib_server(('', 25000))
+```
+
+Our benchmark shall echo the above point,
+
+    $ python threading-server.py
+
+    # In another terminal
+    $ python perf.py
+
+    0.22993946075439453
+    0.22955703735351562
+    0.23864316940307617
+    0.23698139190673828
+
+Compare this with earlier benchmark it’s almost same, To prove you, that it’s
+not concurrent execution, open another window and run `perf.py`
+
+    0.49178433418273926
+    0.4784412384033203
+
+it doubles, now open another terminal and run `perf.py ` it triples and so on…
+
+NOTE: One benefit of using threads is that you could send multiple requests to
+the server(each handled by a thread).Verify this with running server.py in another terminal window(**you can’t send
+more than one request**).
+
+So could we really do something *concurrently*…actually yes… python has got
+something called  and it works by creating separate  that do not share state(for
+a mental model…think of it as separate python interpreters).
+
+But here’s the catch creating is costly(takes more time)…they are not as
+lightweight as threads…but in our case creating new processes would speed up
+things as we have socket -IO which is the slowest portion…
+
+So here’s fully-concurrent code in python,
+
+```python
+# multiprocessing-server code
+
+from fib import fib
+from socket import *
+from multiprocessing import Process
+
+def fib_server(address):
+    sock = socket(AF_INET, SOCK_STREAM)
+    sock.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
+    sock.bind(address)
+    sock.listen(5)
+
+    while True:
+        client, addr = sock.accept()
+        print("Connection", addr)
+        workers = [Process(target=fib_handler, args=(client,)) for i in range(4)]
+        for p in workers:
+            p.daemon = True
+            p.start()
+
+def fib_handler(client):
+    while True:
+        req = client.recv(100)
+        if not req:
+            break
+        n = int(req)
+        result = fib(n)
+        resp = str(result).encode('ascii') + b'\n'
+        client.send(resp)
+    print("Closed")
+
+# This starts a server that listens to localhost on port 25000
+fib_server(('', 25000))
+```
+
+    # Again do the measurement:
+    0.26950716972351074
+    0.272355318069458
+    # it remains almost unchanged with increasing no. of new connections
+
+#### CONCLUSION:
+
+Now when someone says is good at concurrency what they mean is that it has got
+some special constructs which are lightweight and do not share memory( they are
+and  ). So I guess you do get some idea of what concurrency is…But this is not
+even scratching the surface. Following are some links if you want to dive
+deeper,
+
+1.  For state of concurrency in python
+[[here](http://www.dabeaz.com/usenix2009/concurrent/Concurrent.pdf)]
+1.  David Beazley's talk
+[[here](https://www.youtube.com/watch?v=MCs5OvhV9S4&t=896s)] introduces `concurrent-futures` module.
+1.  Rob Pike’s talk “Concurrency is not
+parallelism”[[here](https://www.youtube.com/watch?v=cN_DpYBzKso)]
+1.  go-lang’s concurrency constructs
+[[here](https://www.google.co.in/url?sa=t&rct=j&q=&esrc=s&source=web&cd=3&cad=rja&uact=8&ved=0ahUKEwjxwMbU9t7VAhUBN48KHYLHCwMQFgg0MAI&url=https://www.golang-book.com/books/intro/10&usg=AFQjCNFxHb4i6MKa5YoNAqwV-ujLNMENUw)]
+1.  erlang’s concurrency model
+[[here](http://erlang.org/doc/getting_started/conc_prog.html)][AT YOUR OWN RISK]
+
+Hey amidst all this we forgot our goal *world-domination*, yup doing things
+parallelly/concurrently is TOUGH :)
+
+NOTE: There might be a follow-up article with concurrency implemented in some other language.
+
+
+
+
+
+[1]: The [original
+paper](http://www.cs.utexas.edu/users/EWD/transcriptions/EWD04xx/EWD476.html)
+[somewhat readable].
+
+[2]: So what’s the point in mentioning them? Just to **reiterate** the point
+that this IS one of the **most complex problems** of computer science and there
+have been many approaches to solve
+it[[here](https://www.wikiwand.com/en/Concurrent_computing)]
+
+[3]: There are actual/REAL issues with the way python is designed…IT DOES NOT
+SUPPORT CONCURRENCY BY DESIGN…At the time of its creation(in 1994) it was
+designed for single cores and it has (quite infamous) GIL (Global Interpreter
+Lock) the lock which makes it inherently single threaded.