Bug 165221 – IndexedDB 2.0: Send operations off to the server in batches instead of one at a time

• Component: WebCore Misc.
• Status: RESOLVED FIXED

[Brady Eidson]

IndexedDB 2.0: Send operations off to the server in batches instead of one at a time

Turns out timer scheduling and/or dispatching to the runloop is expensive.

On a profile of "100,000 writes to an IDB", timer scheduling/firing presented as 10+% of the time.

This patch negates much of that.

[Brady Eidson]

Created attachment 295767 [details]
Patch v1

[Brady Eidson]

GTK build breakage is not because of this patch.

[Myles C. Maxfield]

Comment on attachment 295767 [details]
Patch v1

View in context: https://bugs.webkit.org/attachment.cgi?id=295767&action=review

> Source/WebCore/Modules/indexeddb/IDBTransaction.cpp:412
> +    static const size_t operationBatchLimit = 128;

IIRC Ben had an argument against const variables being static. I can't remember it but I think it was about how static variables have to live in their own memory region and therefore have to be a load rather than an immediate.

> Source/WebCore/Modules/indexeddb/IDBTransaction.cpp:-417
> -            schedulePendingOperationTimer();

I assume you are scheduling the operation for the next chunk somewhere else?

[Brady Eidson]

(In reply to comment #3)
> Comment on attachment 295767 [details]
> Patch v1
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=295767&action=review
> 
> > Source/WebCore/Modules/indexeddb/IDBTransaction.cpp:412
> > +    static const size_t operationBatchLimit = 128;
> 
> IIRC Ben had an argument against const variables being static. I can't
> remember it but I think it was about how static variables have to live in
> their own memory region and therefore have to be a load rather than an
> immediate.

"An expert" needs to figure this out for sure, because as is there are 5958 results in 1264 files searching for "static const"

> > Source/WebCore/Modules/indexeddb/IDBTransaction.cpp:-417
> > -            schedulePendingOperationTimer();
> 
> I assume you are scheduling the operation for the next chunk somewhere else?

Correct!

Thanks.

[WebKit Commit Bot]

Comment on attachment 295767 [details]
Patch v1

Clearing flags on attachment: 295767

Committed r209157: <http://trac.webkit.org/changeset/209157>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.