Book FSFS notes

[Greg Hudson <gh...@MIT.EDU>]

I just noticed the current book text:

> On the other hand, FSFS writes the latest version of a file as a
> delta against the next oldest version (modulo some tricks)

Only 50% of file revs are written out as deltas against the
next-oldest version.  So the "(modulo some tricks)" hedge isn't really
enough here; that's a bit like saying the BDB filesystem "stores file
revisions in plain text (modulo some tricks)".

> FSFS also has a much longer delay when finalizing a commit, which
> might cause clients to time-out when waiting for a response.

I feel like this statement exaggerates the corresponding point in my
propaganda page (by adding an unjustified "much"), when field
experience has shown that the point is largely theoretical.  The
finalization delay is usually not noticeable.

> This might also lead to scalability problems when a large number of
> commits are happening simultaneously.  Commit processes may line up
> waiting for a write lock, rather than doing work in parallel
> database transactions.

This point also turned out to be mostly theoretical.  People have
tried to produce it in practice and failed.  I think the volume of
commits involved would cause BDB to completely roll over and die, due
to limitations of its transaction system.

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org