Tickle your brain, I may...

Scenario:

- VERY busy Splunk Syslog server dropping some UDP packets
- Bandwidth is not saturated at the port
- No I/O, CPU, or other resources on the server are saturated

How do I determine the cause of the drops? Limitiation of OS, NIC?

Solution Options?

- Add a second NIC with a unique IP, split up connections on client side
- Add a second NIC and NIC team (bond) them with the same IP
- Run a bunch of VMs with virtual NICs

Please advise!


Regards,

Install Wire Shark and watch the packets. You may get a good idea of what is happening from that. Your Linux distribution may have Wire Shark in its software repository. If so then install from there. If your distribution doesn't have Wire Shark in it software repository then you can get it here.

http://www.wireshark.org/

UDP packets can be dropped if they don't come in in sequence. In theory, if your Splunk server isn't keeping up with the flow, it may cause the packets to become queued up in the buffer. If that happens, you'll get sequence problems, and packets will start to drop.

As UDP doesn't use ACKs, it can be difficult to evaluate congestion problems to the sender, which may skew your reporting slightly.

Try contacting support@splunk.com for more help.

Splunk technical reps were on site just before I posted this. Of coarse their suggestion is more servers, but I find that a bit bothersome as this server is so under-utilized on all other fronts such as CPU, memory, etc..

a quick netstat -su shows:

Udp:
18684705 packets received
2388 packets to unknown port received.
1326491 packet receive errors
1658804 packets sent

The problem of coarse is that this isnt really a Splunk issue, its more a networking issue, so they don't have much else to offer as for advice.

What I am trying to gather some additional knowledge about is a) determine what exactly is causing the udp packet receive errors and b) determining a solution to resolve that.

Any additional help would be greatly appreciated!


Regards,