Not a larger percentage, no.
I just did an experiment with the load testing software.
I had it run a 500-user session with users logging in 5 seconds apart. The 4-CPU machine didn't go crazy--but the CPU was used up pretty well and the LA hit about 8-12.
Then I changed the parameters to 500 users logging in 10 seconds apart. The machine, and the result is:
last pid: 10163; load averages: 1.02, 0.88, 0.94 17:05:51
76 processes: 74 sleeping, 2 on cpu
CPU states: 69.3% idle, 28.8% user, 1.9% kernel, 0.0% iowait, 0.0% swap
Memory: 16G real, 14G free, 324M swap in use, 29G swap free
PID USERNAME LWP PRI NICE SIZE RES STATE TIME CPU COMMAND 9832 httpd 1 59 0 16M 14M sleep 0:21 1.92% httpd 9549 httpd 1 29 0 15M 13M sleep 0:33 1.91% httpd 10112 httpd 1 59 0 14M 12M sleep 0:04 1.56% httpd 9942 httpd 1 19 0 15M 14M sleep 0:13 1.45% httpd 9874 httpd 1 59 0 14M 12M sleep 0:14 1.39% httpd 9700 httpd 1 59 0 15M 14M sleep 0:25 1.39% httpd 10053 httpd 1 59 0 14M 13M sleep 0:08 1.20% httpd 10155 httpd 1 55 0 14M 12M sleep 0:01 1.18% httpd 9848 httpd 1 52 0 14M 12M sleep 0:17 1.17% httpd 9316 httpd 1 59 0 15M 14M sleep 0:41 1.14% httpd 10097 httpd 1 59 0 14M 12M sleep 0:04 0.99% httpd 10156 httpd 1 59 0 13M 11M sleep 0:00 0.93% httpd 10113 httpd 1 29 0 14M 13M cpu/2 0:03 0.88% httpd 10158 httpd 1 59 0 14M 12M sleep 0:00 0.86% httpd 9833 httpd 1 59 0 14M 12M sleep 0:23 0.74% httpd
It's all in the login!
John Madden wrote:
Just as an example: I'm simulating 500 users on this 4-CPU sparc. Here's the top output: 166 processes: 159 sleeping, 3 running, 4 on cpu CPU states: 12.7% idle, 82.7% user, 4.7% kernel, 0.0% iowait, 0.0%
What does it look like if you simulate, say, 10 users? Do your httpd's end up consuming larger percentages of CPU time?
John
begin:vcard fn:Lesli St. Clair n:St. Clair;Lesli email;internet:[EMAIL PROTECTED] version:2.1 end:vcard
