608156 - Run WeaveCrypto in Chrome Worker

Status: RESOLVED WONTFIX

(Firefox :: Sync, defect)

Comment 1

[Doug Turner (:dougt)]

Next time please clone!

570619 blocked b2 / b8 it was marked wontfix and this bug was filed.  Bringing over the same blocking flags.

Comment 2

[Mark Finkle (:mfinkle) (use needinfo?)]

Philipp - assigning to you so it's assigned to someone

Comment 3

[Stuart Parmenter]

Can someone please provide data on how much time we are spending on doing crypto for some profile (and how many records are being decoded?) so we can determine if we actually need to block on this?

Comment 4

[Richard Newman [:rnewman]]

(In reply to comment #3)
> Can someone please provide data on how much time we are spending on doing
> crypto for some profile (and how many records are being decoded?) so we can
> determine if we actually need to block on this?

We're planning on doing some timing work -- Bug 624485, IIRC. Numbers for crypto will be part of this. I'll tackle this after I'm done with MP/wizard work, assuming nothing else drags me away.

Comment 5

[Tony Chung [:tchung]]

(In reply to comment #4)
> (In reply to comment #3)
> > Can someone please provide data on how much time we are spending on doing
> > crypto for some profile (and how many records are being decoded?) so we can
> > determine if we actually need to block on this?
> 
> We're planning on doing some timing work -- Bug 624485, IIRC. Numbers for
> crypto will be part of this. I'll tackle this after I'm done with MP/wizard
> work, assuming nothing else drags me away.

bug 624485 isnt flagged to block firefox or fennec final, so please indicate if that should be.  otherwise, i think the blocking priority on this bug will need to change if its dependent on the other.

Comment 6

[Richard Newman [:rnewman]]

(In reply to comment #5)

> > We're planning on doing some timing work -- Bug 624485, IIRC. Numbers for
> > crypto will be part of this. I'll tackle this after I'm done with MP/wizard
> > work, assuming nothing else drags me away.
> 
> bug 624485 isnt flagged to block firefox or fennec final, so please indicate if
> that should be.

It shouldn't be. The timing numbers are just to make people happy and indicate areas for further work.

> otherwise, i think the blocking priority on this bug will need
> to change if its dependent on the other.

I wouldn't call it a formal dependency; more 'sibling'.

Comment 7

[Richard Newman [:rnewman]]

Added some preliminary instrumentation results in Bug 624485. It looks like NSS operations *and their callers*, all the way up the stack, take no more than 10% of the total time of a sync, and there are many more opportunities for improvement prior to using workers.

Comment 8

[Mike Connor [:mconnor]]

From the results outlined in bug 624485 comment 1, this is likely more trouble than it's worth.  Because we decrypt one record per call, we're talking about thousands of small calls, none of which are long enough on their own to block the main thread for any meaningful amount of time. Without a bulk API in NSS, using ChromeWorker for these ops likely add even more overhead with little benefit.

Instead, we're going to focus on making the crypto pieces much more efficient with the intent of reducing overall time spent in crypto.