[RC5] Active subspaces/random subspaces

R. Kelley Cook kcook at ibm.net
Thu May 7 12:33:36 EDT 1998


On Wed, 6 May 1998 11:45:00 -0500 (CDT), Jim C. Nasby wrote:

>The basic reason for this is that the clients are coded to use the
>current subspace+1 as the random subspace. It might be more space
>efficient to issue out of random keyspaces (ie: switch from 0x64 to 0x65, 
>instead of 0x66), but it would also increase the dupe rate durring
>switchover periods. Also, redoing random subspaces right now isn't an
>option, since that would make a completed subspace the random one, meaning
>that almost all randomly generated blocks would be dupes.
>
>Moo!
>dB!

How about make the random subspace to be current_subspace+2.  And then
when the current subspace is complete to increment by one.  This would
make more sense to me...

Examples:

Current 0x64 -- Random is working on 0x66

When 0x64 is completely checked out.

Current switches to 0x65 which implies that random will 0x67. 
Obviously no time savings over the current system.  Of course, the
random clients that have not checked in will still be working on 0x66
but there will be no overlap here.

Now when 0x65 is completely checked out

Current keyspace becomes 0x66 which then can neatly step around the
sections that the first random round used.  Randoms of course will now
switch to block 0x68.  Hopefully by this time all of the randoms from
block 0x66 will have checked in.  If not, there will be some dupes, but
only moderately more than the randoms clashing within themselves.

I am pretty sure that this scheme will not require any more server
space than the 4 keyspaces that are currently tracked.



--
To unsubscribe, send 'unsubscribe rc5' to majordomo at lists.distributed.net
rc5-digest subscribers replace rc5 with rc5-digest



More information about the rc5 mailing list