Re: Data Storage Issue (Basic Issue)
On Sun, 11 May 2008, Eric Sosman wrote:
Lew wrote:
Eric Sosman wrote:
No. "Suggestive," perhaps, or even "cautionary." But for
my own part I wouldn't confer "very indicative" on a lab result
that cannot be reproduced by an independent experimenter. A
citation of a full report or published paper or something of that
nature might change my mind, but at the moment ... no.
OK, that's fine. It was something that convinced me that it worked in that
case because I could interact with people who conducted the test. You
choose to reject the anecdote because I have no credibility with you.
No, no, no, Lew: Your credibility is just fine (or my credulity; take
it either way). All I'm saying is that the information you've presented
is insufficient to support the inference Tom Anderson drew from it
("kind of damning for filesystems"), and that he should consider being a
little less hasty in drawing sweeping conclusions from sketchy and
incomplete reports.
I'm not saying filesystems are worthless. Just that when it comes to
shunting data on and off a disk, the fact that a different method can work
five times faster than a filesystem means that if you're in the business
of shunting data on and off a disk as fast as possible, a filesystem is
probably not a great idea.
Of course, if there was some particular reason why there was such a big
difference in this case that doesn't apply in all or most cases, then fair
enough, i'm over-generalising. I can't think of what that might be,
though.
If someone told you there was a "five-to-one difference" between
ArrayList and plain array, would you conclude that this was "kind of
damning" for Collections? Of course not. I'm urging Tom not to make
the analogous error.
If someone (credible) told me that there was a five-to-one difference in
times for an ArrayList.get and an array[], then, if we were in a context
where performance was important, i certainly would take that as kind of
damning for collections. I think that's perfectly reasonable.
tom
--
It's almost over now.