using a low interval with icmpping

[bbrendon]

I'm trying to ping a router every 14 seconds. I have the interval set for 14, but when I look at the data, (last 500 values) it shows that the item is running every 59-61 seconds. Is there a way to ping something more agressively or do I need an external script for this?

Thanks!

[Tractor]

Such thing will need some tuning. I haven't checked any deep things, but seems that setting PingerFrequency=3 or similar very small value in zabbix_server.conf will help you.

Anyway, I strongly suggest you to use SNMP traps instead of pinging every few seconds. There are physical, mathematical and technological reasons why ping probe in most cases should be ran not more than one time per 5 minutes - when runing more activelly it will do more errors, and with pinging period less than 10-20 seconds you will be able to monitor that router without errors only in the same lan. In case you need smaller periods of time I once again suggest you to use SNMP traps.

[bbrendon]

PingerFrequency appears to ping the host as a zabbix process and not an item.

SNMP traps won't work for this. I need to see if this internet router is up or not and I won't receive a trap if its down.

[alj]

I'm trying to ping a router every 14 seconds. I have the interval set for 14, but when I look at the data, (last 500 values) it shows that the item is running every 59-61 seconds. Is there a way to ping something more agressively or do I need an external script for this?

Thanks!

14 seconds doesnt sound like too often. The CPU load monitor for unix_t runs every 5 seconds and it works.

Try to check your queue (Monitoring->Queue) if you see there some nonzero numbers in some lines besides first couple ones then your queue is backed up and a bunch of monitors are probably delayed.

To resolve this problem either increase number of zabbix_server daemons in zabbix_server config (i run 80, default is like 5 or so) or disable all hosts that are down at the moment.


For example if i have 5 or more hosts down with Unix_t template assigned my queue backs up forever (those tcp connections waiting for timeout are using up all zabbix_server daemons).

I did not find solution of that yet, if you make number of daemons too big then your database performance goes down and it becomes bottleneck (and queue backs up), if you make it small then your queue will be backed up when some servers go down due to zannix server daemons waiting for tcp timeout.

[Alexei]

I'm trying to ping a router every 14 seconds. I have the interval set for 14, but when I look at the data, (last 500 values) it shows that the item is running every 59-61 seconds. Is there a way to ping something more agressively or do I need an external script for this?

Thanks!

This is because ZABBIX uses only one process for pinging which is unsufficient for large number of servers. This will be improved in 1.4.

[Tractor]

Bit explanation of what is too often:

1. To monitor real times we need to send several pings (does Zabbix really do this?).
2. Ping time can be tens of seconds in some situations (routers overloaded, link overloaded, etc.). I personally had seen ping times of 300 seconds and more (!!!).
3. There should be some additional time to monitoring application itself to process that data, on overloaded machines it can be seconds or even minutes. And does not matter that medium processing time can be hal of second - in this case we need to account peak (biggest) times.
4. And at last, there should be sufficient time gap between two checks to ensure that monitoring server won't get overloaded by itself, e.g., when new check is started while old still processing. Such time should be at least several times bigger than all the time mentioned above, for example, 4 times bigger.

So, if we send three pings with potential delay ~10sec. and provide up to 10 secs to Zabbix itself, it give us 40 seconds of processing time. When you add 4 times bigger gap of free time you will need more than 3 minutes between checks.

So, such small times as 14 seconds are possible only in case when it is totally impossible that ping will be more than 1-2 seconds and monitoring system itself have lots of free processor and RAM resources.

[Vince2]

Zabbix uses fping that can ping several hosts at once and wait for their answers. It takes less 5 seconds to ping many hosts even if most of them are unreachable. Therefore, there is no need to wait 300 seconds or even 40 seconds.

[Tractor]

Vince, you don't understood. This has nothing to do with pinging program itself but with general network and monitoring characteristics. And fping itself solves only monitoring machine resources problem because it helps avoiding fork fork fork :-)

[Alexei]

ZABBIX 1.3.x support configurable number of ICMP pingers.