Re: [Gzz] 25th, 26th, 27th & 28th (hh)

[hemppah]

Quoting Tuomas Lukka <address@hidden>:

> > 
> > As I have said earlier, I don't know Gzz very well yet. That's why it's
> quite
> > hard to answer Gzz specific questions (how Gzz *really* differ from
> existing p2p
> > systems). But I will try to respond to research questions:
> 
> Maybe it would be better first to try to answer all the ones you can and
> only
> then really write. Because this is the major part of the thesis!

Yes, that's true.

> 
> > -For research questions a, b, and c, I know what the p2p field has to
> offer: the
> > best effort is DHTs with O(log n) (with small variations).
> 
> Again: with what assumptions? Who has to be trusted? How reliably is a
> block
> going to be found?

For normal conditions, with n nodes, the DHT algorithm require O(log n) hops to
reach arbitrary destinations *and* each node must mantain information about
O(log n) neighbours. Additionally, DHT is very robust for failures; for example,
when 20% of the nodes simultaneously fail, less than 0.1% of the routes fail.

The biggest problem in DHTs is that there might be hostile nodes which route
requests purposely to wrong directions. However, there are proposal how these
kinds of issues could be solved.


> 
> Does this work also with the "local store" model?

What do you mean by "local store" (replication or something else) ?

>> -For e, there are some proposals, e.g. SQL-like searchs in DHTs and keyword
>> searchs. However, this is quite Gzz specific, e.g. Urn-5 etc. (?)

>Urn-5: think about it as a very unique keyword. 

What do you mean by saying (in research questions): "Mist? tied?n, mik? on
*suht* oikea ?" --> the word "suht".

> > -For f, same algoritmical limits hold true in this too. 
> 
> Which ones?

Now, look answer above.

> 
> > This is also Gzz
> > specific and is, at least someway, implentation issue when we have the
> basic
> > infrastructure builded (a, b, c)
> 
> It is similar to c) but not the same.

That's true, but (of course) same algorithmical limits applies also in this 
case.

> 
> > -For g and h, same as f (?)
> 
> Not a satisfactory answer.

Hmm...in practise, what g and h mean ?

> 
> > -For i, There are some clever implementations for this in Groove. But I
> don't
> > know how it can scale in situations where there are *lot* of users. Some
> other
> > keywords which might be relevant are SDSI and SPKI. However, I haven't put
> much
> > effort on this yet.
> > 
> > If I remember right, a and b was the most important questions, right ?
> 
> They all are important. Maybe you should really try to work on that chapter
> first.

Ok :). Research questions a, b, c are examined ("the best algorithmical effort
from p2p field for routing"), but perhaps I should work more on other questions
also.


-Hermanni


-------------------------------------------------
This mail sent through IMP: http://horde.org/imp/