This document discusses strategies for optimizing Ruby garbage collection performance. By default, Ruby uses a mark and sweep garbage collector and allocates memory in exponentially increasing heap sizes. A better strategy is to stabilize on 1-2 constant heap sizes by setting a high minimum heap size and using linear heap size increments. Various tools like memprof and gdb can provide statistics on memory usage and garbage collection cycles to analyze performance.

1. Ruby Performance
scaling your Ruby web app
Part 1: Garbage collection

2. MRI/REE Facts
• Mark & sweep GC
• freelists
• very expensive
• All refs are on the heap
• All code is on the heap

3. GC Settings
RUBY_HEAP_MIN_SLOTS 10k
RUBY_HEAP_SLOTS_GROWTH_FACTOR 1.8
Defaults
startup
+ 20k
+ 10k
+ 18k
...
10k
10k + 10k
10k + 10k + 18k
10k + 10k + 18k + 32.4k
...
RUBY_HEAP_SLOTS_INCREMENT 10k
10 000
20 000
38 000
70 400
128 720

4. A better strategy
• Stabilize on 1-2 heaps
RUBY_HEAP_MIN_SLOTS (480 000)
• Linear increment to keep it tight
RUBY_HEAP_SLOTS_GROWTH_FACTOR 1
RUBY_HEAP_SLOTS_INCREMENT (20 000)

5. Some tools
memprof, gdb, scrap
[68.63 MB] GET /stats/scrap
Number of objects : 815266 (627150 AST nodes, 76.93%)
Heap slot size : 20
GC cycles so far : 352
Number of heaps : 7
Total size of objects: 15923.16 KB
Total size of heaps : 18036.55 KB (2113.39 KB = 11.72%
unused)
Leading free slots : 18545 (362.21 KB = 2.01%)
Trailing free slots : 0 (0.00 KB = 0.00%)
Number of contiguous groups of 16 slots: 2685 (4.65%)
Number of terminal objects: 4697 (0.51%)

6. Thanks
Gleb Mazovetskiy. github.com/glebm