Will the program of 3h stop the previous one started at 1h?
Emin Gabrielyan
Salut,
le teste est lancé sur astrad5 (176 clients):
astrad5:/etc/astrad/cron# cat /etc/cron.d/astrad
# m h dom mon dow command
30 23 * * * root ntpdate ntp.ovh.net >> /dev/null 2>&1
*/3 * * * * root /etc/astrad/cron/croncheck
*/1 * * * * root /etc/astrad/script/ast-registration.pl >> /dev/null 2>&1
00 01 * * * root /etc/astrad/script/ast-notify-send-1.pl
00 03 * * * root /etc/astrad/cron/teststart
00 05 * * * root /etc/astrad/cron/teststop
|
* * *
On 2011-06-17 13:49, Task-By Emin Gabrielyan wrote:
You need to launch the transmission of notifies one by one
during 3h then replace it with a transmission 20 by 20. The best is to
do this in the middle of the night. You can maybe program this with a
cron, such as, until 1:00 nothing, between 1:00 and 3:00 notifies one
by one, between 3:00 and 5:00 notifies 20 by 20.
Emin Gabrielyan
Salut,
Ci-dessous les graphes d’utilisation CPU pour astrad6 et astrad7 avant
et après l’activation de Notify (durant la nuit : de 00h à 07h)
L’envoi des Notify est regroupé par 10 depuis hier.
J’ai remarqué que l’utilisation du script optimisé (10 par 10) à une
légère amélioration sur l’utilisation CPU mais je n’ai pas suffisamment
de donnée pour prouver cette hypothèse.
On 2011-06-17 02:03, Task-By Emin Gabrielyan wrote:
The second chart shows that notify costs us and is not
completely free. You do not indicate the frequency. The chart without
frequency is not really informative. The CPU of notify sent one by one
must be compared with the CPU of notifies sent in blocks. This will
permit to see if we gain anything by sending in blocks.
Emin Gabrielyan
Salut,
Ci-dessous le CPU durant le taste de notify :
astrad6: notify + ngrep
astrad7: notify.
<moz-screenshot-15.jpg>
On 2011-06-14 14:07, Task-By Emin Gabrielyan wrote:
Did you do this? Try not to invoke the asterisk command for each block of notifies, but rather try to use the same asterisk command and provide to it the input from a file or via a pipe.
Emin Gabrielyan
Le Jun 14, 2011 à 0:41, Emin Gabrielyan <emin.gabrielyan@switzernet.com> a écrit :
Measure the CPU load when sending notifies to 200 clients every 15 seconds. Send 20 notifies in a single command. Do this during 5 minutes and show on the CPU chart, before during, and after.
Emin
Envoyé de mon iPad
|