[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Mon, 14 Sep 2015 07:47:37 +0530
From: Sayantan Datta <std2048@...il.com>
To: john-dev <john-dev@...ts.openwall.com>
Subject: Re: bt.c initial table size
On Sun, Sep 13, 2015 at 1:55 AM, Solar Designer <solar@...nwall.com> wrote:
> Sayantan -
>
> When testing with lots of hashes, I was getting many "Progress is too
> slow!! trying next table size." and this hurt the speeds badly (2 to 3.5
> minutes total running time for my test). To have bt.c use a larger
> table size right away, I hacked this into create_perfect_hash_table(),
> right before the "do { ... } while(1);" loop:
>
> for (i = 0; i < 9; i++) {
> unsigned int temp;
> temp = next_prime(approx_offset_table_sz % 10);
> approx_offset_table_sz /= 10;
> approx_offset_table_sz *= 10;
> approx_offset_table_sz += temp;
>
> i++;
>
> if (!(i % 5)) {
> multiplier_ot += inc_ot;
> multiplier_ht += inc_ht;
> approx_offset_table_sz = (((long
> double)num_loaded_hashes / 4.0) * multiplier_ot + 10.00);
> approx_hash_table_sz = ((long
> double)num_loaded_hashes * multiplier_ht);
> }
> }
>
> The loop contents is some lines I copied from your larger loop. Oh, I
> just noticed I got an extra "i++" in there. Well, I am writing this to
> ask you to make something like this a standard feature, where we could
> request a larger initial table size without hacking the source code -
> and maybe the default needs to be bumped up too?
>
> Alexander
>
Hi Solar,
How would you like to request the table size - the exact table size or a
multiplier(as in table size = num_loaded_hashes * multiplier + prime) .
Also, if you request a larger table size, you are not guaranteed that it
will 'build the tables on first try', however, the probability of that
happening is definitely higher.
Regards,
Sayantan
Content of type "text/html" skipped
Powered by blists - more mailing lists
Confused about mailing lists and their use?
Read about mailing lists on Wikipedia
and check out these
guidelines on proper formatting of your messages.
