Compute the heap growing factor based on mutator utilization and allocation throughput.

Compute the heap growing factor based on mutator utilization and allocation throughput.

Doc: https://goo.gl/LLGvBs

BUG=

Committed: https://crrev.com/143a9e04312b061061c95f918f99ccc4f2f67541
Cr-Commit-Position: refs/heads/master@{#29015}

[ulan, 2015-06-05 18:29:28]

ulan@chromium.org changed reviewers:
+ hpayer@chromium.org

[ulan, 2015-06-05 18:29:28]

Hannes, wdyt?

I will add gc-tracer function to report the full incremental marking duration
(total steps + final pause) per GC. Until then this works only with
--noincremental-marking.

[ulan, 2015-06-08 07:48:27]

> I will add gc-tracer function to report the full incremental marking duration
Turns out we don't need the full marking duration and can use the existing
marking speed and mark-compact speed. I'll update the CL.

[Hannes Payer (out of office), 2015-06-08 21:21:43]

I like it.

It is not really mutator utilization, because it just depends on allocation
throughput. Let's call it mutator allocation utilization or something similar.

Let me know when it is ready for review.

[ulan, 2015-06-10 13:56:47]

Hi Hannes, this is ready for review, ptal.

> It is not really mutator utilization, because it just depends on allocation
throughput. Let's call it mutator allocation utilization or something similar.
As discussed offline,  the function also depends on the target mutator
utilisation (mutator-time / (gc-time + mutator-time)) to compute the heap
growing factor assuming that the given allocation throughput stays the same
until next GC.

Should I changed the description to
"Compute the heap growing factor based on mutator utilization and allocation
throughput"?

[Hannes Payer (out of office), 2015-06-11 01:33:57]

I like it. Just a few minor comments.

Yes, please update the description.

https://codereview.chromium.org/1163143009/diff/210001/src/heap/gc-tracer.cc
File src/heap/gc-tracer.cc (right):

https://codereview.chromium.org/1163143009/diff/210001/src/heap/gc-tracer.cc#...
src/heap/gc-tracer.cc:622: const double kEpsilon = 0.5;
Can we rename epsilon to kMinimumMarkingSpeed (or something more descriptive)?

https://codereview.chromium.org/1163143009/diff/210001/src/heap/heap.cc
File src/heap/heap.cc (right):

https://codereview.chromium.org/1163143009/diff/210001/src/heap/heap.cc#newco...
src/heap/heap.cc:5305: // https://goo.gl/LLGvBs.
Instead of posting a link, can we add the necessary details here? I am fine with
spending several lines on comments.

https://codereview.chromium.org/1163143009/diff/210001/src/heap/heap.cc#newco...
src/heap/heap.cc:5310: if (max_old_generation_size_ <=
kMaxOldSpaceSizeMediumMemoryDevice) {
If we would not access max_old_generation_size_, this function would be
unit-testable? WDYT? Having tests for the function would be nice.

[ulan, 2015-06-11 08:49:38]

Thanks for comments.

I changed the description and uploaded new patch set.

https://codereview.chromium.org/1163143009/diff/210001/src/heap/gc-tracer.cc
File src/heap/gc-tracer.cc (right):

https://codereview.chromium.org/1163143009/diff/210001/src/heap/gc-tracer.cc#...
src/heap/gc-tracer.cc:622: const double kEpsilon = 0.5;
On 2015/06/11 01:33:57, Hannes Payer wrote:
> Can we rename epsilon to kMinimumMarkingSpeed (or something more descriptive)?

Done.

https://codereview.chromium.org/1163143009/diff/210001/src/heap/heap.cc
File src/heap/heap.cc (right):

https://codereview.chromium.org/1163143009/diff/210001/src/heap/heap.cc#newco...
src/heap/heap.cc:5305: // https://goo.gl/LLGvBs.
On 2015/06/11 01:33:57, Hannes Payer wrote:
> Instead of posting a link, can we add the necessary details here? I am fine
with
> spending several lines on comments.

Done.

https://codereview.chromium.org/1163143009/diff/210001/src/heap/heap.cc#newco...
src/heap/heap.cc:5310: if (max_old_generation_size_ <=
kMaxOldSpaceSizeMediumMemoryDevice) {
On 2015/06/11 01:33:57, Hannes Payer wrote:
> If we would not access max_old_generation_size_, this function would be
> unit-testable? WDYT? Having tests for the function would be nice.

Done. The first function to be unit-tested in Heap :)

[Hannes Payer (out of office), 2015-06-12 05:18:30]

Did you forget to add the tests?

[ulan, 2015-06-12 07:26:14]

On 2015/06/12 05:18:30, Hannes Payer wrote:
> Did you forget to add the tests?
Ah, yes, uploaded the missing test file.

[Hannes Payer (out of office), 2015-06-15 06:57:38]

lgtm

[ulan, 2015-06-15 08:01:11]

The CQ bit was checked by ulan@chromium.org

[commit-bot: I haz the power, 2015-06-15 08:01:26]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1163143009/250001

[commit-bot: I haz the power, 2015-06-15 08:32:55]

Committed patchset #14 (id:250001)

[commit-bot: I haz the power, 2015-06-15 08:33:09]

Patchset 14 (id:??) landed as
https://crrev.com/143a9e04312b061061c95f918f99ccc4f2f67541
Cr-Commit-Position: refs/heads/master@{#29015}