Increase timeouts when using a 1000 hz. From the FreeBSD commit message: ÊChange TCPTV_MIN to be independent of HZ. ÊWhile it was documented to

Êbe in ticks "for algorithm stability" when originally committed, it turns
Êout that it has a significant impact in timing out connections. ÊWhen we
Êchanged HZ from 100 to 1000, this had a big effect on reducing the time
Êbefore dropping connections.

ÊTo demonstrate, boot with kern.hz=100. Êssh to a box on local ethernet
Êand establish a reliable round-trip-time (ie: type a few commands).
ÊThen unplug the ethernet and press a key. ÊTime how long it takes to
Êdrop the connection.

ÊThe old behavior (with hz=100) caused the connection to typically drop
Êbetween 90 and 110 seconds of getting no response.

ÊNow boot with kern.hz=1000 (default). ÊThe same test causes the ssh session
Êto drop after just 9-10 seconds. ÊThis is a big deal on a wifi connection.

ÊWith kern.hz=1000, change sysctl net.inet.tcp.rexmit_min from 3 to 30.
ÊNote how it behaves the same as when HZ was 100. ÊAlso, note that when
Êbooting with hz=100, net.inet.tcp.rexmit_min *used* to be 30.

ÊThis commit changes TCPTV_MIN to be scaled with hz. Êrexmit_min should
Êalways be about 30. ÊIf you set hz to Really Slow(TM), there is a safety
Êfeature to prevent a value of 0 being used.

ÊThis may be revised in the future, but for the time being, it restores the
Êold, pre-hz=1000 behavior, which is significantly less annoying.

ÊAs a workaround, to avoid rebooting or rebuilding a kernel, you can run
Ê"sysctl net.inet.tcp.rexmit_min=30" and add "net.inet.tcp.rexmit_min=30"
Êto /etc/sysctl.conf. ÊThis is safe to run from 6.0 onwards.

ÊApproved by: Êre (rwatson)
ÊReviewed by: Êandre, silby

1 files changed, 9 insertions, 0 deletions

diff --git a/etc/rc.bootup b/etc/rc.bootup
index 75e1590..07108d2 100755
--- a/etc/rc.bootup
+++ b/etc/rc.bootup
@@ -316,6 +316,15 @@
 		mwexec("sysctl net.inet.tcp.recvspace=4096");
 		mwexec("sysctl net.inet.tcp.sendspace=4096");
 	}
+	
+	/* if we are operating at 1000 then increase timeouts.
+	   this was never accounted for after moving to 1000 hz */
+	$kern_hz = `sysctl kern.clockrate | awk '{ print $5 }' | cut -d"," -f1`;
+	$kern_hz = trim($kern_hz, "\r\n");
+	if($kern_hz == "1000") {
+		mwexec("sysctl net.inet.tcp.rexmit_min=30");
+		mwexec("sysctl net.inet.tcp.rexmit_min=30");		
+	}
 
 	upnp_start();