[rc5] Random keyblocks

Sanford Olson sanford at msn.fullfeed.com
Tue Aug 26 15:15:37 EDT 1997


At 01:36 PM 8/26/97 -0600, Eric Gindrup wrote:
[snip]
>        The real question is a question of utility.  The most utile thing 
>     to do, until the probability of finding the key in a tardy block is 
>     larger, is to assign unassigned blocks to clients.  When it is more 
>     likely that a tardy block contains the target key, i.e. when the 
>     number of tardy blocks is greater than twice the number of unassigned 
>     blocks, it is more utile to work on tardy blocks.
[snip]

I have to disagree here.  A tardy block is never more likely to contain a
key than an unassigned block.  As a group, when the number of tardy blocks
exceed the number of unassigned blocks (by a 2 to 1 margin assuming the
average tardy block was checked half-way), then, yes, the target key is
more likely to be in the tardy block group than the unassigned group.  But,
we are not checking "groups", we are checking blocks, so even if there is
only one unassigned block left, it is more likely to contain the target key
than any given tardy block.  Thus, we should to the most efficient thing
and check all the unassigned blocks first before worrying about the tardy
ones.  All these ideas of caching or buffering tardy blocks are pointless -
if you are going to cache anything, it might as well be unassigned blocks.
As far as making random blocks more efficient, I don't think that random
blocks are a significant portion of checked blocks and I don't think anyone
should be wasting time improving the algorithms for random blocks - the
time could be better spent improving either client speed, reliability, or
network communications.

- Sanford
----
To unsubscribe, send email to majordomo at llamas.net with 'unsubscribe rc5' in the body.



More information about the rc5 mailing list