[3796] in Kerberos
Re: Max. 'size' of Kerberos IV database
daemon@ATHENA.MIT.EDU (Mike Friedman (510) 642-1410)
Sat Sep 3 13:27:06 1994
To: kerberos@MIT.EDU
Date: Sat, 03 Sep 94 09:46:36 PDT
From: spgmnf@cmsa.Berkeley.EDU (Mike Friedman (510) 642-1410)
In article <BRLEWIS.94Sep3101123@joy.mit.edu>
brlewis@joy.mit.edu (Bruce R. Lewis) writes:
>
>If I understand it correctly, ndbm doubles the virtual size of the
>database whenever two keys have the same hash. Your expect script is
>probably producing many principal names with the same hash value. Try a
>different algorithm for naming your 20000 principals (e.g. pseudo-random
>strings).
>--
Very insightful! In fact, I first created a file containing principal/
password pairs, generated by a perl script. Then, my expect script used
that file as input. I used a simple naming convention for the principals:
prnnnnn, with nnnnn ranging from 00001 to 20000. Indeed, I suppose this
would tend to create hashing collisions!
I can try a different algorithm for generating principal names. Of
course, I hope than our real principal names (whatever they turn out
to be) will be randomly distributed enough to help out significantly.
Meanwhile, I intend to pursue the suggestion I received from a couple of
people to use another dbm-compatible package that doesn't create such
relatively large sparse files.
Thanks.
