Michal Ochman ocher

Latency numbers every programmer should know

L1 cache reference ......................... 0.5 ns
Branch mispredict ............................ 5 ns                     on recent CPU
L2 cache reference ........................... 7 ns                     14x L1 cache
Mutex lock/unlock ........................... 25 ns
Main memory reference ...................... 100 ns                     20x L2 cache, 200x L1 cache
Compress 1K bytes with Zippy ............. 3,000 ns  =   3 µs
Send 2K bytes over 1 Gbps network ....... 20,000 ns  =  20 µs
SSD random read ........................ 150,000 ns  = 150 µs

Read 1 MB sequentially from memory ..... 250,000 ns = 250 µs 4X memory

Proper use of html_safe

Let's look at an innocuous piece of ruby. Consider some view code showing a user's name and phone number:

"#{first_name} #{last_name} #{phone}"

Great - this is very succinct, readable, and can easily be extracted to a method in a

	$ jruby -v
	jruby 1.7.5.dev (1.9.3p392) 2013-09-04 090d5dd on Java HotSpot(TM) 64-Bit Server VM 1.7.0_25-b15 [darwin-x86_64]
	$ time jruby -J-XX:+TieredCompilation -J-XX:TieredStopAtLevel=1 -J-Xverify:none -e 'require "rails"'
	jruby -J-XX:+TieredCompilation -J-XX:TieredStopAtLevel=1 -J-Xverify:none -e 5.18s user 0.47s system 136% cpu 4.142 total
	$ time jruby -e 'require "rails"'
	jruby -e 'require "rails"' 14.60s user 0.44s system 202% cpu 7.444 total

	# Thee will be more information here when I share the entire problem space I'm working on, but
	# in short, this is preview material for my second talk in a series called "What Computer Scientists Know".
	# The first talk is on recursion, and goes through several examples., leading up to a problem based
	# on a simple puzzle that initial estimates based on performance of a previous puzzle would take years
	# to solve on modern computers with the techniques shown in Ruby. That sets the stage for improving the
	# performance of that problem with threading, concurrency, and related tuning.
	#
	# The second talk is on threading and concurrency, touching on algorithmic performance as well.
	# Using some knowledge of the problem (board symmetry, illegal moves, etc), we reduce the problem space
	# to about .5% of what we initially thought it was. Still, the initial single threaded solution took more

	var app = angular.module('myApp', []);

	/* Set up a simple controller with a few
	* examples of common actions a controller function
	* might set up on a $scope. */
	app.controller('MainCtrl', function($scope, someService) {

	//set some properties
	$scope.foo = 'foo';
	$scope.bar = 'bar';