684581 - Investigate 2k chunks

Status: RESOLVED INCOMPLETE

(Core :: JavaScript: GC, defect)

Description

[Igor Bukanov]

Consider the following page:

<script>
alert(1);
</script>

When it shows the alert about:memory gives for it after I pressed minimize memory usage on 32-bit Linux:

69,632 B (00.09%) -- gc-heap
│  │      │  ├──40,960 B (00.05%) -- arena-unused
│  │      │  ├──15,208 B (00.02%) -- objects
│  │      │  ├──11,520 B (00.01%) -- shapes
│  │      │  ├─────896 B (00.00%) -- type-objects
│  │      │  ├─────392 B (00.00%) -- arena-padding
│  │      │  ├─────384 B (00.00%) -- scripts
│  │      │  └─────272 B (00.00%) -- arena-headers

That is, among almost 70K of GC aenas we waste 40K or 59% in arena tails.

Now consider another page:

<script>
var a = 1;
</script>

After it loads I have:

110,592 B (00.10%) -- gc-heap
│   │     │  ├───47,872 B (00.04%) -- arena-unused
│   │     │  ├───32,680 B (00.03%) -- shapes
│   │     │  ├───27,808 B (00.02%) -- objects
│   │     │  ├────1,152 B (00.00%) -- type-objects
│   │     │  ├──────432 B (00.00%) -- arena-headers
│   │     │  ├──────392 B (00.00%) -- arena-padding
│   │     │  └──────256 B (00.00%) -- scripts 


here again we waste over 40% in arena tail. It also shows way too much memory allocated for shapes, I file that in another bug.

This tells for pages with small amount of JS code 4K arenas are not suitable. So we should investigate 2K arenas and see how it would affect the memory usage.

Comment 1

[Nicholas Nethercote [inactive]]

If we didn't have so many different arena kinds the problem would also presumably be smaller.

Comment 2

[Andrew McCreight [:mccr8]]

Probably we don't need a 14 year old speculative bug.