On 21.08.2013 21:11, Navdeep Parhar wrote:
I'd like to add a last-active timestamp to the structure that tracks the LRO state in a NIC's rx handler. This is r254336 in user/np/cxl_tuning that will be merged to head if there are no objections.
No objections. This is good thing. The last time I looked some time back there were a couple of additional issues with the soft-LRO code. One was that it would cause packet re- ordering when a new segment is out-of-order or can't be merged. There the new packet would be delivered first and the merged LRO chain after it. I think there were a couple of other similar issues. However I don't know if these have been fixed in the mean time. BTW an excellent resource on the correct LRO behavior is this detailed flow chart. It probably would be good to audit our soft-LRO against it. http://msdn.microsoft.com/en-us/library/windows/hardware/jj853325%28v=vs.85%29.aspx -- Andre
http://svnweb.freebsd.org/base?view=revision&revision=254336 http://svnweb.freebsd.org/base/user/np/cxl_tuning/sys/netinet/tcp_lro.c?r1=254336&r2=254335&pathrev=254336 http://svnweb.freebsd.org/base/user/np/cxl_tuning/sys/netinet/tcp_lro.h?r1=254336&r2=254335&pathrev=254336 ----- Add a last-modified timestamp to each LRO entry and provide an interface to flush all inactive entries. Drivers decide when to flush and what the inactivity threshold should be. Network drivers that process an rx queue to completion can enter a livelock type situation when the rate at which packets are received reaches equilibrium with the rate at which the rx thread is processing them. When this happens the final LRO flush (normally when the rx routine is done) does not occur. Pure ACKs and segments with total payload < 64K can get stuck in an LRO entry. Symptoms are that TCP tx-mostly connections' performance falls off a cliff during heavy, unrelated rx on the interface. Flushing only inactive LRO entries works better than any of these alternates that I tried: - don't LRO pure ACKs - flush _all_ LRO entries periodically (every 'x' microseconds or every 'y' descriptors) - stop rx processing in the driver periodically and schedule remaining work for later. ----- Regards, Navdeep _______________________________________________ freebsd-net@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-net To unsubscribe, send any mail to "freebsd-net-unsubscr...@freebsd.org"
_______________________________________________ freebsd-net@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-net To unsubscribe, send any mail to "freebsd-net-unsubscr...@freebsd.org"
