Update: This MAY be caused by net.inet.tcp.rfc6675_pipe=1, which I had recently enabled since I am now moving data over WAN-style distances on a regular basis and since both ends are FreeBSD with this capability I figured it might help performance.
If so it almost-certainly isn't ARM-specific...... I shut that off (it was recently turned on) and thus far I'm not able to recreate the network hangs.... Hmmmm... CCing into freebsd-net, as this is looking increasingly like a non-ARM-specific thing. Feel free to remove freebsd-arm from responses (in fact, probably ought to unless someone can confirm it's working properly on other architectures but not on ARM.) On 5/12/2017 12:45, Karl Denninger wrote: > Under fairly heavy stress (~50% of the 100Mbps possible FDX performance) > I've now run into a problem that is turning into something I can repeat > without too much trouble by transmitting a large file through the device. > > The system appears to "crash" from the outside. It's not dead, however > -- it's got some sort of network buffer hang going on. This is a Pi3 > with current (-HEAD, FreeBSD 12.0-CURRENT #0 r318193M: Thu May 11 > 16:18:20 CDT 2017) software on it -- reverting to a mid-March kernel did > NOT change the behavior. > > The symptomology is that the unit will start printing this on the console: > > May 12 12:25:46 IPGw dhcpd: send_packet: No buffer space available > May 12 12:25:46 IPGw dhcpd: dhcp.c:3974: Failed to send 300 byte long > packet over ue0.3 interface. > > ue0.3 is a VLAN for a private subnet and is not specific to the issue; > it also complains about no space on the primary too: > > May 12 12:27:06 IPGw dhcpd: dhcp.c:3974: Failed to send 300 byte long > packet over ue0 interface. > May 12 12:27:13 IPGw dhcpd: send_packet: No buffer space available > > netstat -m does /not /show any denied or delayed requests for network > buffers or mbuf exhaustion. Systat -vm shows plenty of RAM available > (roughly half.) > > An ifconfig ue0 down / ifconfig ue0 up sequence from the console clears > the hang. > > Has anyone else seen anything similar to this? > -- Karl Denninger k...@denninger.net <mailto:k...@denninger.net> /The Market Ticker/ /[S/MIME encrypted email preferred]/
smime.p7s
Description: S/MIME Cryptographic Signature
