[JDEV] performance numbers.. questions

[temas]

What you want to start looking at is the messages/sec that are being
pushed through the server this is what is going to be hitting the CPU
hard.

--temas

On 09 May 2001 14:52:20 -0500, Dustin Puryear wrote:
> I am actually now getting some hard numbers from the jabbertest tools.
> One test I ran connected 1000 users to a Jabber server running on a
> Pentium III 600 with 192MB of RAM. I saw some odd results. But first,
> here is how the test works:
> 
> active_jab basically connects 1..n users serially, and sends x messages
> when each user connects. It does NOT send messages concurrently for each
> connected user. (Although that would be an excellent feature--to be
> added soon!) Instead, if you want to simulate multiple users sending
> messages at the same time to the same server then you just run muliple
> instances of active_jab on the same or different client machines. (Very
> simple.) The basic idea is:
> 
> for user 1..n do
>   connect user to jabber
>   send x messages and track delivery times
> done
> 
> for user 1..n do
>   disconnect user from jabber
> done
> 
> Memory
> =======
> First of all, I assumed that Jabber would consume a lot of memory, which
> turned out to be a wrong assumption. I saw my %memused jump from a rough
> minimum of 81% to maybe 95%. My %swpused did not change significantly,
> which indicates that I had enough core memory for the job. (Remember
> Linux always uses as much memory as it can for buffers, so the starting
> value of 81% isn't surprising.) The big issue here is that there was
> only a small jump--around 14%.
> 
> CPU
> =======
> Big surprise here. Jabber seems to *quickly* begin eating CPU cycles as
> the number of connected users increases. Here is some data:
> 
> users   idle  user  sys
> 0       99%   1%    0%
> 200     70%   24%   3%
> 500     45%   50%   5%
> 800     30%   64%   6%        
> 1000    20%   70%   10%
> 
> (Yes, the totals may be more or less that 100%. I am getting these
> numbers from graphs and quickly jotting them down.)
> 
> Note that Jabber is consuming 24% of the CPU at 200 users on a Pentium
> III 600. I also noticed that as more users are connected the message
> delivery time decreased. The progession was slow.. usually going from 0
> seconds to 2 seconds as the number of users went from 0..1000. But
> still.. why is that? Remember I am not sending message concurrently.
> Instead, as each user connects active_jab sends x messages (defaults to
> 5), and then proceeds to the next user.
> 
> I would like some feedback on this data. Does this look like what
> everyone else has been experiencing? Are my numbers out of wack?! What
> other numbers would be useful?
> 
> I think the next logical step is to run active_jab on several machines
> at once. That way we can test how many *concurrent* users can connect
> and send messages.
> 
> Regards, Dustin
> 
> -- 
> Dustin Puryear <dpuryear at usa.net>
> http://members.telocity.com/~dpuryear
> In the beginning the Universe was created. 
> This has been widely regarded as a bad move. - Douglas Adams
> _______________________________________________
> jdev mailing list
> jdev at jabber.org
> http://mailman.jabber.org/listinfo/jdev