At 4:26 -0700 1/10/2000, Daniel de Rauglaudre wrote:
>> >From what I understand about the type of distributed search used by EDCL,
>> there is an arbitrary selection for distinguished points so that the number
>> of points that the server must handle is manageable. It seams to me that
>> you could cut the average time for the search in half by adding a second
>> server.
>
>You mean multiply the number of distinguished points by two? We could
>do that, but we would not accelerate the process.
>
>Actually, we dont compute on million distinguished points but one
>million of billion points validated by one million distinguished
>points. Each point "represent" on average one billion points.
>
>We could compute two millions distinguished points but the number of
>points to compute would remain one million of billion points! In these
>case, we would just obtain the distringuished points twice faster, but
>each of them "representing" (on average) half a billion points instead
>of one billion points.
>
>Therefore if we need e.g. 5 hours to compute a distinguished point, we
>would save 2.5 hours in the total time!
Ahh, the light turns on! The pseudo random path is generated from the value
of the current point so that once a match is generated the two series will
stay in sync until the match is detected by reaching a distinguished point.
Thanks for clarifying this.
PS: is it ever possible for a series to loop and never reach a
distinguished point?
-- Dan Oetting <oetting@ghtmail.cr.usgs.gov>
This archive was generated by hypermail 2b29 : Mon Jan 10 2000 - 17:30:22 MET