503760 - Measure deflated string cache

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Andreas Gal :gal]

The deflated string cache has a serious finalization cost in the GC. We should measure what benefits we get out of it at runtime in practice.

Comment 1

[Andreas Gal :gal]

Proposed measurement approach:

- Measure hit ratio and total browser memory use with the current code.

- Disable the cache (i.e. make it always miss) and then measure total browser memory use again.

We should also measure how often we hit js_DeflateString in general, and how much time we spend inside of it.

Comment 2

[Brendan Eich [:brendan]]

(In reply to comment #1)
> Proposed measurement approach:
> 
> - Measure hit ratio and total browser memory use with the current code.
> 
> - Disable the cache (i.e. make it always miss) and then measure total browser
> memory use again.

Do you know what you're proposing here? "Disable the cache" must leak returned string byte vectors, since there is no mandatory JS_PutStringBytes API. So you will measure the leaks accumulating, not merely the loss of sharing due to the disabled cache.

> We should also measure how often we hit js_DeflateString in general, and how
> much time we spend inside of it.

It would be great to measure lifetime of strings from which deflated bytes are accessed.

Note that we already use a DEFLATED flag bit in the JSString header to avoid the costly finalization. Are you sure you're not on a snipe hunt?

Also note bug 491579 especially bug 491579 comment 10.

/be

Comment 3

[Luke Wagner [:luke]]

This has all been removed.