From: Jay V. <fu...@us...> - 2012-12-21 05:31:23
|
WOOLLEY Dan <Dan...@uk...> wrote: >I am using the latest 2.1.4 e1000e intel Linux driver with the Linux >boning driver on RHEL 5.4 >(http://www.kernel.org/doc/Documentation/networking/bonding.txt) > >If I set the miimon value to 50 milliseconds, everything works okay, >but the link down detection using netif_carrier_ok() is always about 350 >msec. This is too long for our application. > >Do you know of any way to decrease this detection latency ? > >I can use the arp_monitoring instead, but there are other issues doing >this (increase traffic and false positives etc). > >I have also tried the "use_carrier=0" option which makes it use the >deprecated MII or ETHTOOL ioctls, but it makes no difference (still 350 >msecs). Presumably you're not setting the "downdelay" option to bonding (which will cause this effect deliberately). You don't say what chipset you're using, but my next guess is the chipset and/or driver has some delay in the carrier loss detection. Bonding's detection of carrier loss is ultimately dependent on the chipset for notification, so there's probably nothing you can tweak in bonding to change this. What chipset do you have? I don't think e1000e prints that out at probe time, so you'd have to look it up with lspci, e.g., on my e1000e "lspci | grep Ethernet" gives me: 00:19.0 Ethernet controller: Intel Corporation 82567LM Gigabit Network Connection (rev 03) and then "lspci -vvn -s 00:19.0 gives me: 00:19.0 0200: 8086:10f5 (rev 03) Subsystem: 17aa:20ee [... a bunch more stuff ...] Kernel driver in use: e1000e and then hopefully somebody from Intel can tell you what sort of link loss detection time should be expected from your driver / chipset combination (which may be different from what I have). -J --- -Jay Vosburgh, IBM Linux Technology Center, fu...@us... |