stavro / viterbi

Accompanying code for a USC presentation

Tweet Tracker

This repository accompanies a presentation made for USC Course 31934, Spring 2016: Web Application Project.

Goal: Examine a traditional three tier web application compared to a stateful web service.

Contained within this repository are two API servers capable of tracking and persisting occurrences of hashtags, and responding with a global statistics counter.

• Three-Tier: A traditional stateless web app delegating all stateful behavior to the database.
• Stateful: A stateful web app leveraging in-process data locality to respond immediately to requests.

Both servers have two routes exposed:

• GET /: Retrieves statistics on all tracked hashtags
• POST /track?hashtag=<name>: Tracks a usage event of a particular hashtag.

Both servers are backed by a Postgres instance.

Instructions:

• Clone repository, create the database, and ensure both servers can start.
• Benchmark both servers for reading and writing.
• (extra credit) - Generate 100k+ rows and repeat the benchmark.
• Share and discuss results.

Database

Both apps use the same database. Once you have Postgres installed, create a database called tweets:

1. createdb tweets
2. psql -d tweets
3. create table tweets (id serial primary key, hashtag text not null, created_at timestamp not null);

Starting your web server

1. cd three_tier (or cd stateful)
2. bundle install
3. bundle exec ruby app.rb
4. verify the app loads by going to http://localhost:4567/ on your browser

Benchmarking

Once you have a web server running, time to benchmark!

Install wrk: brew install wrk

# GET '/'
wrk -d10s --timeout 2000 http://localhost:4567/

# POST '/track?hashtag=<insert hashtag>'
wrk -d10s --timeout 2000 -s track.lua http://localhost:4567/

(Extra Credit) Generating Extra Rows

insert into tweets (hashtag, created_at) select 'hashtag_' || (s.i / 100)::text, now() as created_at from generate_series(1,100000) as s(i);