Garbage Collection and the Ruby Heap

The talk will start out with a brief overview of stacks vs heaps, and various garbage collection algorithms. We’ll explain the details of the MRI implementation of ruby heap (10k initial slots, grows by 1.8x each time, linked list of free slots) and garbage collector (stop the world mark-and-sweep collector, scans the C stack, triggered by 8mb malloc_limit and rb_newobj).

After explaining the basics, we’ll focus on specific examples and show off a variety of tools and patches:

• understanding and tuning the GC: ObjectSpace statistics, RAILSBENCH

• finding reference leaks in ruby: BleakHouse, GC.dump_heap

• reducing memory usage and overhead: Copy-on-Write GC / REE, tcmalloc

• finding C extension memory leaks: valgrind, google-perftools heap profiler

We’ll also highlight common tricks for improving the performance of ruby code running under MRI by reducing the memory overhead:

• store commonly used strings in constants instead of using them inline
• use symbols instead of strings
• invoke GC after large allocations
• disable GC during time sensitive operations

Finally we’ll look ahead to ruby 1.9 and jruby and talk about how they fix some of these issues in MRI.