[ruby-core:93696] [Ruby master Feature#15997] Improve performance of fiber creation by using pool allocation strategy.

Issue #15997 has been updated by ioquatix (Samuel Williams).


@ko1 asked:

> (1) stack size assumption

The fiber pool stack size is (guard page + vm_stack_size + fiber_machine_stack_size).

> (2) maximum allocatable size

On 64-bit platform it's effectively the same, although in some situations it can be better due to reduced number of `mmap`s required.

On 32-bit platform, it's slightly worse, because I didn't bother implementing fallback on `mmap` failure. In current implementation, worst case difference is 128 fiber stacks. That being said, if you are allocating fibers up to the limit of the 32-bit address space you will quickly run into other issues, so I don't consider this a bug, it's just natural limit of 32-bit address space.

> (3) GC.enable/disable usage (edited) 

- `vm2_fiber_allocate` is running with `GC.disable` to do fair comparison of allocation overheads.
- `vm2_fiber_count` is running with normal GC, but due to using alloca on fiber pool stack, GC pressure/count is significantly reduced. It is not expected to represent expected improvement of real world code, but shows that fiber pool code in isolation avoids GC overheads.
- `vm2_fiber_reuse ` is running with `GC.disable` and deterministically calls `GC.start` after allocating 1024 fibers to test performance of fiber reuse.
- `vm2_fiber_switch` is existing benchmark and is not affected by fiber pool implementation.


----------------------------------------
Feature #15997: Improve performance of fiber creation by using pool allocation strategy.
https://bugs.ruby-lang.org/issues/15997#change-79314

* Author: ioquatix (Samuel Williams)
* Status: Open
* Priority: Normal
* Assignee: ko1 (Koichi Sasada)
* Target version: 
----------------------------------------
https://github.com/ruby/ruby/pull/2224

This PR improves the performance of fiber allocation and reuse by implementing a better stack cache. As per @ko1's request, we also increased fiber stack size to be the same as thread stack size.

The fiber pool manages a singly linked list of fiber pool allocations. The fiber pool allocation contains 1 or more stack (typically more, e.g. 512). It uses N^2 allocation strategy.

```
//
// base = +-------------------------------+-----------------------+  +
//        |VM Stack       |VM Stack       |                       |  |
//        |               |               |                       |  |
//        |               |               |                       |  |
//        +-------------------------------+                       |  |
//        |Machine Stack  |Machine Stack  |                       |  |
//        |               |               |                       |  |
//        |               |               |                       |  |
//        |               |               | .  .  .  .            |  |  size
//        |               |               |                       |  |
//        |               |               |                       |  |
//        |               |               |                       |  |
//        |               |               |                       |  |
//        |               |               |                       |  |
//        +-------------------------------+                       |  |
//        |Guard Page     |Guard Page     |                       |  |
//        +-------------------------------+-----------------------+  v
//
//        +------------------------------------------------------->
//
//                                  count
//
```

The performance improvement depends on usage:

```
Calculating -------------------------------------
                     compare-ruby  built-ruby 
  vm2_fiber_allocate     132.900k    180.852k i/s -    100.000k times in 0.752447s 0.552939s
     vm2_fiber_count       5.317k    110.724k i/s -    100.000k times in 18.806479s 0.903145s
     vm2_fiber_reuse      160.128     347.663 i/s -     200.000 times in 1.249003s 0.575269s
    vm2_fiber_switch      13.429M     13.490M i/s -     20.000M times in 1.489303s 1.482549s

Comparison:
               vm2_fiber_allocate
          built-ruby:    180851.6 i/s 
        compare-ruby:    132899.7 i/s - 1.36x  slower

                  vm2_fiber_count
          built-ruby:    110724.3 i/s 
        compare-ruby:      5317.3 i/s - 20.82x  slower

                  vm2_fiber_reuse
          built-ruby:       347.7 i/s 
        compare-ruby:       160.1 i/s - 2.17x  slower

                 vm2_fiber_switch
          built-ruby:  13490282.4 i/s 
        compare-ruby:  13429100.0 i/s - 1.00x  slower
```

Additionally, we conservatively use `madvise(free)` to avoid swap space usage for unused fiber stacks. However, if you remove this requirement, we can get 6x - 10x performance improvement in `vm2_fiber_reuse` benchmark. There are some options to deal with this (e.g. moving it to `GC.compact`) but as this is still a net win, I'd like to merge this PR as is.





-- 
https://bugs.ruby-lang.org/

Unsubscribe: <mailto:[e-mail]?subject=unsubscribe>
<http://lists.ruby-lang.org/cgi-bin/mailman/options/ruby-core>