Array Slicing fo’ Shizzle!: Array Slicing fo’ Shizzle! © Kris 2007
Some Background: Some Background 2001: Web servers (Tomcat, Resin, WebLogic, etc)
2002: Bull in a china shop
A Java Enigma: A Java Enigma Overt GC reliance has potential to make any language or system appear sluggish
Java has few easy-to-apply alternatives
Libraries & systems can appear GC-bound
GC can be a synchronization bottleneck
Yet, Java flourished on the server
GC Pressure Reduction: GC Pressure Reduction Low-level access to malloc/free
Precision new & delete overrides
Temporal struct or class on the stack
Free-lists
Array slicing
Array Slicing fo’ Shizzle!: Array Slicing fo’ Shizzle! Easy-to-use array aliasing
Native to the language
auto slice = array [start .. end];
Three Use-Cases: Three Use-Cases Text Processing
Sliced HTTP
Clustering with Tina
1: Integer as Text: 1: Integer as Text char[] uitoa (uint val, char[] output=null)
{
if (output.length < unit.dig)
output.length = uint.dig;
auto len = output.length;
do {
output [--len] = val % 10;
} while (val /= 10);
return output [len .. $];
}
1: UTF Conversion: 1: UTF Conversion dchar[] toUtf32 (char[] input, dchar[] output=null)
{
if (output.length < input.length)
output.length = input.length;
// convert input utf8 into output utf32
// …
return output [0 .. len];
}
1: Parsing Lines of Text: 1: Parsing Lines of Text // option 1 via an array
foreach (line; splitLines (myText))
… The quick brown fox jumped over the lazy dog
How now brown cow
The rain in Spain falls mostly on the plain // option 2 via opApply
foreach (line; lines (myText))
…
2: Sliced HTTP: 2: Sliced HTTP Client
2: Sliced HTTP Server: 2: Sliced HTTP Server Cookies Headers URI Request structured alias input buffer overlap optional
2: Thread Local Storage: 2: Thread Local Storage
3: Clustering Model: 3: Clustering Model Bulletin Task Queue Cache
3: Model Services: 3: Model Services Bulletin Task Queue Cache put (msg)
get () put (key, msg)
get (key, remove) execute (msg) broadcast (msg)
3: Tina: 3: Tina High performance
Peer to peer design (no SPOF)
Serialized D objects
Near zero GC activity
3: Storage Request: 3: Storage Request Client Object buffer
3: Task Request: 3: Task Request
Tango Connection: Tango Connection Text processing supports slicing
Conversion
Tokenizing
Streaming
Utilities
IO subsystem supports slicing
Protocol readers (allocators)
Buffering
Wrap-up: Wrap-up Overt GC reliance has potential to make any language or system appear sluggish
Array slicing can be a powerful alternative
D libraries & systems can easily be GC-lite
Less multi-thread contention
Show some GC love
Thank You: Thank You