Re: [lvs-users] Handling Duplicate ACKs, and a Game of Pong.

To: chchen@xxxxxxx
Subject: Re: [lvs-users] Handling Duplicate ACKs, and a Game of Pong.
Cc: lvs-users@xxxxxxxxxxxxxxxxxxxxxx
From: Charlie Allom <charlie@xxxxxxxxxxxxxx>
Date: Sun, 27 Mar 2011 12:58:23 +0100
On Fri, Mar 25, 2011 at 01:10:27PM -0700, chchen@xxxxxxx <chchen@xxxxxxx> wrote:
> Did you have a symptom in general? Mine was SSL problems with
> Windows 7 machines, and duplicate ACKs for those handshakes.


Yes my symptom is a rogue ACK/FIN/RST (same seq/ack #) that is sent multiple
times from director -> backupdirector, and then the backupdirector ping
pongs that packet on its loopback interface at up to 800Mbps and brings
down the machine with interrupts.

And of course the 3rd real server with no LVS enabled doesnt have this
problem (bare metal eth0 aliases).

So the issue is really the primary director fucking up, and /really/
confusing the backup. I'd like to revert the 2.0.2 bnx2 driver in newer
kernels.. but changing to LVS-NAT may be simpler, as you suggested.

I don't know if that means I lose the ability of the directors to be
realservers or not.

Also I wonder if there is some sysctl I could tweak, as the loopback
garbage doesn't scream bnx2 problems.


Please read the documentation before posting - it's available at: mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
Send requests to lvs-users-request@xxxxxxxxxxxxxxxxxxxxxx
or go to

<Prev in Thread] Current Thread [Next in Thread>