Re: random users on threads

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

Hi Jeremy,

On 7/11/07, Jeremy Brown <jb...@no...> wrote:
> The problem I have is that I need to test a large numbers of users

Great, I like it!
Y can see some recent staff devoted to it here:
http://curl-loader.sourceforge.net/high-load-hw/index.html

> with fewer IP's(threads) then users.

Not understanding this.
Current implementation enables either to use a unique IP-address
for user/client or to use a single common IP-address (which limits
the theoretical number of users/clients to 64K - number of ports to bind()).

All users are doing in a single thread now, unless we are not
separating them to some N number by using -t <N> option.
Thus, lets say when 20 000 users/clients are run with -t 4 option
are separated to 4 sub-batches, each of 5000 users/clients, and
each sub-batch runs independently in a separate thread, where
the first (0) thread collects all statistics.

> I would like to use a random user each time in the thread runs. ie 100,000 users and 400 threads

Threads are quit another concept, and normally lower numbers, e.g. equal to
to the numbers of logical CPUs are recommended.

>I already have a file with the users in the FORM_RECORDS_FILE and
would like to just >grab a new one each time the thread runs.

Would you like to run each cycle a one new user or several new users?
Lets say from a pool of 100 000 users to run each cycle a 1000 users?

> Or I could use >UNIQUE_USERS_SAME_PASSWORD with the userX with the same password and just >enable random on the users for each iteration/cycle.

I got it. There is no currently support for running some random users,
but patches are welcomed!
In any case, we will add your proposal to our TODO list.

-- 
Sincerely,
Robert Iakobashvili,
coroberti %x40 gmail %x2e com
...........................................................
http://curl-loader.sourceforge.net
A web testing and traffic generation tool.