when snmp trap packets are sent at rate of (1000/min.)
and more, to a host running snmptrapd, it is found that
snmptrapd drops the trap packets. The number of
packets being dropped increases further as the rate
increases.
Is there any way to optimize the snmptrapd.c to
receive traps generated at the rate of (1000-8000/min.) ?
Logged In: YES
user_id=93926
What version of our software are you using?
What platform are you on? (i.e. what does 'uname -a' report?)
Logged In: YES
user_id=567057
What version of our software are you using?
What platform are you on? (i.e. what does 'uname -a' report?)
Logged In: YES
user_id=567057
I am using net-snmp version 5.0.1. And platform that I am
using is GNU/Linux Red Hat 7.2 (kernel 2.4.10).
Logged In: YES
user_id=88647
The first question is: is this cpu-bound or slowed down by
other factors? What does top say about the box?
One obvious bottleneck would be resolving ip addresses to
host names. Are you running it with -n option?
Logged In: YES
user_id=726199
How can you test this problem?
Can you send me the test suites and the test steps? I am
really interested in this issue and want to take a look.
I can be reached at dpan817@yahoo.com
Logged In: YES
user_id=658204
I wonder how to make snmptrapd generate a trap counts it
received?
seems no such a argument.
Logged In: YES
user_id=658204
I use a script file as the snmptrapd handler and get count
from it.
Here's my result
parameter
traps/min 1489 2971 1605 2976
received traps 1489 2188 911 1006
lost traps 0 783 694 1970
on localhost
with -n parameter without -n parameter
traps/min 2975 2977
received traps 2975 2977
lost traps 0 0
seem snmptrapd can handle about 1500-2000 traps/min for a
traphandler
Logged In: YES
user_id=848638
snmptrapd is single-threaded and can only deal with one
notification at a time. Until the notification is processed,
the server socket receive buffer will have to hold the
packet. You can enlarge this buffer in 5.2.x and later by
using the serverRecvBuf config directive (see snmp.conf(5)).
Also, consider using INFORMs instead of TRAPs, because the
sender will resend the packet if it hasn't been
received/acknowledged. There's nothing more we can/will do
to this problem as long as our toolkit isn't threaded.
Hi,
i am facing similer issue where i need to receive 600 to 800 trap/sec. i am working on windows 7 machine and want to implemet the above mention setting for windows machine.
i have installed net-snmp 5.6.1.1
1) what are the configuration i need to do for receiving up to 800 traps /sec without any trap getting drop
2) what is the defaut serveRecvBuf size ?
3) i am using select and snmp_sess_read api to read pdu from socket is there any way to make this multi threaded?
please help me out with this query as i am not able to fugire out how can this be achived
Thanks in advance.
Regards,
Santosh Yadav