Re: url probability bug fix and random number use update

[Robert I. <cor...@gm...>]

Hi Val,

On Sun, Mar 14, 2010 at 8:54 AM, Robert Iakobashvili <cor...@gm...>wrote:

> Hi Val,
>
> On Mon, Mar 8, 2010 at 7:16 AM, Val Shkolnikov <va...@nv...> wrote:
>
>> Hi Robert,
>>   There is a major bug in the curl-loader regarding use of the
>> FETCH_PROBABILITY tag.  In loader_fsm.c, svn rev 574, line 1359 is
>>
>> cctx->url_curr_index =  (size_t) url_next;
>>
>> The url index is updated but not the url pointer.  The result of this bug
>> is that all urls are selected regardless of the probability somewhat
>> uniformly.
>>
>> While debugging this I also reworked the use of probability in the
>> curl-loader to produce better pseudo-random numbers and also added a tag
>> RANDOM_SEED that lets you control the seed.  I am attaching the patch
>> against svn rev 574.  To use it
>>
>> cd curl-loader
>> patch -p3 <curl-loader.patch.574+vs6
>>
>
> Applied the patch, thanks!
>
>>
>> I also noticed that the tag FETCH_PROBAILITY_ONCE is not used, but will
>> let you decide whether it's worth to keep.
>>
>
> Please, look at the latest version in svn. Thanks!
>
> The array of url_fetch_decision was allocated per client only, when
> FETCH_PROBAILITY_ONCE was defined.
> The array is used to cache the fetching decision and to decrease
> the calls to random() for high-load testing.
>
> In most of the cases, however, it will be of no use, and it was corrected
> not to use url_fetch_decision array, when it is not supposed to be
> allocated.
>
>

Have you a chance to look at the version in svn and try the changes made?
Thanks!

-- 
Truly,
Robert Iakobashvili, Ph.D.
......................................................................
www.ghotit.com
Assistive technology that understands you
......................................................................