This post is about a naive benchmark: NodeJS + Cassandra VS Ruby on Rails + MySQL. There are plenty of benchmarks in the wild(db/web). However, it’s still difficult to decide which stack solution to use for your application. Actually, it heavily depends on the application/use case/scenario. This benchmark is based on a very simple scenario: a click count application. This benchmark is very naive! Be aware of that when you use it as a reference.

这篇帖子写得是关于NodeJS + Cassandra VS Ruby on Rails + MySQL的性能对比。网上其实有很多很多这样的比较（数据库/网站框架）。但是我看了以后还是觉得很难决定用哪个产品来做哪个应用。大家都论调不同，视角不同，比较的人也很难说对所有的产品都精通，难免有偏颇，甚至有的可能有广告嫌疑。反正，性能的对比很大程度上依赖于应用/案例/场景。我的这个简单的比较是基于一个简单的点击计数的案例。（这个比较非常初步，仅供参考，可能以后会更深入的研究一下）

NodeJS + Cassandra VS ROR + MySQL

Setup

• Both use an individual virtual machine (VirtualBox) with 1 core cpu & 2 GB Memory
• Install the stack in their respective vm
• Start the services (NodeJS, Cassandra, Ruby on Rails, MySQL) with no tunning - Use configuration come with default, minium modification to make it up & running
• Granularity is hour

Application Implementation

NodeJS + Cassandra

• Stack: Cassandra, node cassandra driver, Express, NodeJS, nvm, npm

• DB Schema (CQL):

1

CREATE TABLE count_by_widget ( widgetID int, time timestamp, clicks counter, PRIMARY KEY (widgetID, time) ) WITH CLUSTERING ORDER BY (time DESC)

• Query to insert a hit (CQL):

1

UPDATE count_by_widget SET clicks = clicks + 1 WHERE widgetID=? AND time=?

• Application Code: For detailed code, see my Github

• Command to start server:

1

NODE_ENV=production node bin/www

ROR + MySQL

• Stack: MySQL, mysql2 gem, Rails, Ruby, rvm, bundler

• DB Schema (ActiveRecord Migration):

1
2
3
4
5
6
7
8
9
10
11
12

class CreateClicks < ActiveRecord::Migration
  def change
    create_table :clicks, {id: false} do |t|
      t.integer :widgetID
      t.timestamp :time
      t.integer :click

      t.timestamps
    end
    execute 'ALTER TABLE clicks ADD PRIMARY KEY (widgetID, time);'
  end
end

• Query to insert a hit (ActiveRecord):

1
2
3
4
5
6

class Click < ActiveRecord::Base
  def self.hit(wid)
    time = Time.new.beginning_of_hour
    Click.create(widgetID: wid, time: time, click: 1) if 0 == (Click.where(widgetID: wid, time: time).update_all 'click = COALESCE(click, 0) + 1')
  end
end

• Application Code: For detailed code, see my Github

• Command to start server:

1

bundle exec passenger start -d -e production

Execution

To generate load, I used my python script. It send requests to a target url with multiple threads. I chose 3 urls associated with three click target ID’s. The rule is simple, push the applications to the limit. If it can handle more, create more threads to generate more load until reach the limit.

Result

NodeJS crashed when RPM goes to around 36,000

Conclusion

Be aware that the data set is not large

• NodeJS + Cassandra can handle more requests. I think Cassandra is faster to perform counter increment.

• NodeJS + Cassandra response the request faster, because it’s non-blocking. Requests can be responsed while db query is still running in NodeJS.

• Ruby on Rails + MySQL is more stable. It didn’t crash when it’s overloaded. Maybe I didn’t implement the NodeJS correctly? Please comment.

• NodeJS + Cassandra is much easier to develop for this use case. e.g. Insert & Update are same in cassandra. That means you don’t have to create a new row before the click increment query.

• NodeJS + Cassandra retreives data faster, since the data is sorted by timestamp in storage. On the other side, MySQL requires scanning over the entire table and then sort.

• Cassandra has a better scalability. By adding Cassandra nodes, performance can be improved in a linear fashion.

Future Study

This benchmark is very naive. The following points can be improved in the future.

• Larger data set

• Mix & Match: NodeJS + MySQL Ruby on Rails + Cassandra

• More DB: MongoDB CouchDB

• More web framework: PHP Django

• Reading operation performance

Reference

• Cassandra VS MongoDB
• Couchbase VS Cassandra the two links above are really conflicting
• Rails VS NodeJS nodejs的吞吐量貌似没有它说的那么牛，但是他对rails的优势劣势分析得还是挺中肯的
• Web App Framework Benchmarks
• Cassandra
• NodeJS
• Express
• Node Cassandra Driver