RE: [Madeleine-devel] Using buffered writes to avoid one fsync per transaction

[Stephen S. <ste...@ma...>]

> Which makes Batched about twice as fast for me - did
> you mean the other way around? Or maybe we have found
> a problem with Batched (in Linux/Windows)?

I had them the right way around.  But the tests are running so much faster
on my machine (nice new twin xeon raid box) that I suspect the results are
ok.  The disk IO is so fast on my box that batching doesn't help (or maybe
it would with a reduced delay).

> Another "problem", maybe Windows reletad, seen above,
> is that I get different number of assertions everytime
> I run ThreadedStartupTest. This happens regardless of
> which persister that's used. It seems like it's this
> assert that's not always "counted":
> 
> assert_equal(n, mad[n].system.z, "object change
> mad[#{n}].z")

I see that on unix too.  It's almost certainly because Test::Unit is not
thread safe.  I've been meaning to submit a patch for it, but it requires
some effort and thought to do right.

Stephen