gps very kindly ran an FHR analysis job for me. It pulled down 6.3M documents 
from non-release users (because the release dataset is not currently reliable 
due to dupes).

This is essentially our beta audience. I don't anticipate that the release 
channel would be substantially different, but it could be either way -- either 
because beta users are heavier users, or because release users typically have 
longer-term hoarded profiles.


Only 8K documents were unable to provide a count of bookmarks.


Here's the raw distribution:

> quantile(y, c(.25, .50, .75, .85, .95, .98, .99, .995, .999))
  25%   50%   75%   85%   95%   98%   99% 99.5% 99.9%
   15    21    34    50   159   413   763  1326  4342


A stock Firefox installation has 23 bookmark entries. If we skip these -- that 
is, if we discount users who don't appear to use bookmarks at all:

> users <- y[y > 23]
> quantile(users, c(.25, .50, .75, .85, .95, .98, .99, .995, .999))
     25%      50%      75%      85%      95%      98%      99%    99.5%   99.9%
  30.000   38.000   69.000  121.000  388.000  871.000 1503.000 2576.000 7988.606



So depending on whether you agree with that last step, a solution that hits 99% 
of our users must support 1503 (or 763) bookmark records. To hit 99.9% (i.e., 
to address ~400M users and not address half a million) requires support for 
closer to 4000/8000 records.

Of course, most of these records won't be changing a lot -- such is the nature 
of bookmarks -- which implies that incrementality will be important.

The vast majority of our users will be around or below 100 bookmarks.


Later today I'll run a similar analysis for Places as a whole, which will 
inform history storage.

If I have time, I'll compare these numbers to our known Sync stats, which will 
profile that population.

My intuition is that a large portion of our install base are not 
"profile-oriented users" -- that is, Firefox for them is mostly a window 
through which they look at content, not a place to accrete data and tools -- 
and that Sync users by definition *are* profile-oriented. But that's largely 
pure speculation.
_______________________________________________
Sync-dev mailing list
[email protected]
https://mail.mozilla.org/listinfo/sync-dev

Reply via email to