Claudio Jeker <cje...@diehard.n-r-g.com> writes: > Don't forget to increase the UDP recvbuffer space. The default is somewhat > small and will result in drops. At least you should invest some time to > play with that value and see if it helps.
I played with the net.inet.udp.recvspace sysctl option today. The maximum allowed value for it is 262144. After that things break (as in no daemons start, network is disabled and during boot I get the message ifconfig: socket: No buffer space available). I guess I am reaching some sort of kernel limit, without producing kernel panic. With the maximum value I do see udp packet drops in my tests (that increase as the test provides larger query rates) root@dmeg-dns1 ~ # netstat -s -p udp udp: 268217 datagrams received 0 with incomplete header 0 with bad data length field 0 with bad checksum 194 with no checksum 267842 input packets hardware-checksummed 0 output packets hardware-checksummed 130 dropped due to no socket 0 broadcast/multicast datagrams dropped due to no socket 0 dropped due to missing IPsec protection 64705 dropped due to full socket buffers ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 203382 delivered 212059 datagrams output 187019 missed PCB cache Unfortunately I see no real difference in BIND's performance with the values I tested (262144, 131072). -- Kostas Zorbadelos twitter:@kzorbadelos http://gr.linkedin.com/in/kzorba ---------------------------------------------------------------------------- () www.asciiribbon.org - against HTML e-mail & proprietary attachments /\