On Mon, Nov 03, 2003 at 11:24:42AM +0000, Ian Clarke wrote:
> Tracy R Reed wrote:
> >On Mon, Nov 03, 2003 at 09:41:23AM +0000, Ian Clarke spake thusly:
> >toad, last week I think. I don't have the IRC logs hanging around.
> 
> Well, toad is big enough and ugly enough to defend his own statements.
> 
> >>By that argument almost none of non-symbolic AI would work since 
> >>frequently it finds solutions to problems which are extremely difficult 
> >>for people to decipher.
> >But people understand how those things work and they know what they are
> >supposed to do and they usually know when they are actually working.
> 
> As do we.
> 
> >There
> >is code in freenet which actively promotes specialization and which routes
> >to other nodes based on their specialization betting that the chances of
> >that node having the data are greater than others.
> 
> To the extent that there is such code, it probably shouldn't be there. 
> It may have belonged in pre-NGR since that did tend to have simple 
> specialization but it didn't have to worry about network latencies and 
> things like.

Hmm. What code is there? Are you talking about pcaching? That is
supposed to enhance routing generally...
> 
> > That stuff does not yet seem to be working.
> 
> What is "working"?  You don't know that most of the current DNFs aren't 
> legitimate, some probably aren't.  If you want to be constructive why 
> not join the discussion about a better way to determine Freenet's 
> performance rather than going on and on about psuccess when we all know 
> that it isn't a good performance measure?

We have a better performance measure, which Tracy suggested and I
implemented. It's called routingSuccessRatioCHK. So far it looks pretty
bad on my node... Of course it's only a better measure if people don't
poll CHKs, but zab tells me that Frost hasn't for two months and most
Frost users upgrade very regularly.
> 
> >Don't get me wrong, I'm patient and I know this stuff
> >is hard. I just don't want anyone to think freenet meets expectations for
> >a functioning network. :)
> 
> "functioning" is pretty objective, Freenet is functioning, but that 
> doesn't mean that it doesn't have much potential for improvement.
> 
> >You almost confused me a second time because I was about to pull up my
> >datastore histogram. You asked me this question on IRC once. I'll tell you
> >the same thing I told you then: Linux is the "CHK" or "key" of the
> >information at the center of my specialization. Logistics and aircraft
> >maintenance is the CHK of my fathers. Networking is the CHK of my
> >roommates. I know what they specialize in and I route to them
> >appropriately when I need something in that area. If nobody specialized in
> >anything in the real world we would have problems. 
> 
> Ok, and do you suppose a histogram of you, your father's, or your 
> roommate's specializations would be clear cut and obvious, or do you 
> suppose that your "common sense" knowledge would probably obscure what 
> you perceive to be your own specialization were it to be drawn as a 
> histogram?
> 
> >No, I'm not. But I bet that when people write code for non-symbolic AI
> >they know how the code is supposed to work and what the end result should
> >look like.
> 
> Yes, and in this case we do too - the end result is that Freenet finds 
> data when it exists, and the accuracy of the routing code can be 
> measured by comparing actual to estimated response times.

I suppose I will implement this eventually. But it will only be useful
for comparing two builds - the variability will be so high that you
can't say a given delta is reasonable.
> 
> > The credit card company expects their code to actually find
> >fraud.  If it didn't appear to turn up fraud and the programmer said, "Oh,
> >this is just non-obvious fraud" he wouldn't be taken very seriously. :)
> 
> Freenet does find stuff.  Freenet doesn't always find stuff.  When 
> Freenet doesn't find stuff it isn't always Freenet's fault because 
> sometimes the stuff doesn't exist.  My last statement does not imply 
> that any time Freenet doesn't find stuff that it isn't Freenet's fault. 
>  Why is this so difficult for you to grasp?
> 
> Anyway, if you want to help stop griping and join the thread about 
> better ways to measure Freenet's performance.

One good way is the latency test that somebody posted here recently.
Insert a file, get one other node to request it. If you can do this in a
single retry 90% of the time, Freenet is working. If you can do it in a
reasonable overall round-trip time, Freenet is working fast.
> 
> Ian.

-- 
Matthew J Toseland - [EMAIL PROTECTED]
Freenet Project Official Codemonkey - http://freenetproject.org/
ICTHUS - Nothing is impossible. Our Boss says so.

Attachment: signature.asc
Description: Digital signature

_______________________________________________
Devl mailing list
[EMAIL PROTECTED]
http://dodo.freenetproject.org/cgi-bin/mailman/listinfo/devl

Reply via email to