MAG(i)C Pseudo-random Number Generator

This pseudo-random number generator generates integers between 1 and 1,000,000. This feature is implemented on two platforms in two different languages, Java and Python on Google App Engine instances and Google Compute VMs.

The software include in this repository meets the following requirements:

• For every 1000 numbers generated, at least 750 of them are unique.
• Numbers are displayed only in Arabic format with no leading zeroes.
• Webpage will display one number at a time in plaintext. To retrieve a new number, the page must be refreshed.

Platform Specific Information

Google App Engine

When implemented on Google App Engine, the site auto-scales to accommodate for the load that the site is experiencing. This is desirable, however be aware that if the site starts experiencing lots of traffic, that it will start to cost more money. It might be a good idea to set a daily spending limit to prevent any problems from arising.

Java AppEngine Instance
Python AppEngine Instance

Google Compute VM

Setup for Google Compute VMs is much more involved than that of App Engine, however it is easier to predict costs and optimize the server as all aspects of the server are configurable.

Java Compute VM Instance
Python Compute VM Instance - Europe
Python Compute VM Instance - US Central

Language Specific Information

Python

The Python version of this app was written using Flask, a micro web-framework that itself is written in Python. Flask was used in order to create a simple and lightweight web experience that is also easily scalable up to very large scale projects.

For some context on how small and lightweight the random number generator site is, here is all of the core functionality of the site:

app = Flask(__name__)

@app.route('/')
def hello():
    return str(random.randint(0,1000000))

Additionally, the Python implementation uses the Web Server Gateway Interface (WSGI) calling convention to handle requests between the flask framework and the web server we are using, Nginx.

Nginx is more than capable of hosting this simple site, but it was selected for its simplicity, flexibility, and excellent performance under high load. Do note that in the App Engine implementation Nginx is not explicitly implemented, but is the default web server for App Engine applications.

Java

The Java version of this app was as a servlet, a native object type in Java EE. Creating a servlet allows the application to be quickly developed and deployed to most java based web servers.

For some context on how simple it is to create a webapp as a servlet, here is the core functionality of the site:

public class RandomEngine extends HttpServlet {

  @Override
  public void doGet(HttpServletRequest request, HttpServletResponse response) 
      throws IOException {

    response.setContentType("text/plain");
    response.setCharacterEncoding("UTF-8");

    Random rand = new Random();
    
    response.getWriter().print(rand.nextInt(999999)+1);

  }
}

Because the webapp was developed using built-in Java EE functionality, we could choose from most java based webservers. We selected TomCat as it provides a straight forward way of adding and maintaining webapps.

Performance

The Python VM version on a Google Compute Engine VM n1-standard-2 instance is capable of handling a little over 3000 requests/sec, and in testing seemed limited only by CPU power. Here is a wrk benchmark on aforementioned instance.

MacBook-Pro:~ andrewm$ wrk -t2 -c400 -d60s --latency http://pyvm.andrewreaganm.me/
Running 1m test @ http://pyvm.andrewreaganm.me/
  2 threads and 400 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency    73.69ms   40.35ms   1.63s    93.79%
    Req/Sec     1.72k   301.03     2.30k    87.73%
  Latency Distribution
     50%   65.09ms
     75%   80.65ms
     90%   99.09ms
     99%  212.33ms
  202983 requests in 1.00m, 34.05MB read
  Socket errors: connect 151, read 0, write 0, timeout 0
Requests/sec:   3378.26
Transfer/sec:    580.25KB

The Java VM version on a Google Compute Engine VM n1-standard-2 instance is capable of handling 6800 requests/sec. Here is a wrk benchmark on aforementioned instance.

Andrews-MacBook-Pro:~ andrewmassey$ wrk -t2 -c200 -d30s --latency http://javm.andrewreaganm.me/
Running 30s test @ http://javm.andrewreaganm.me/
  2 threads and 200 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency    33.20ms   35.04ms   533.80s    96.26%
    Req/Sec     3.46k   750.14     4.13k    87.69%
  Latency Distribution
     50%   26.23ms
     75%   28.33ms
     90%   35.15ms
     99%  215.91ms
  204609 requests in 30.09s, 23.23MB read
Requests/sec:   6800.97
Transfer/sec:    790.83KB