2013/8/8 Jan Friesse <jfriesse at redhat.com>: > ...So my recommendation for you is: > - Upgrade to latest corosync (depend on stream you are running, so ether > 1.4.6 or 2.3.1) I'm already running 1.4.6. > - Some switches don't like fragmented packets and this can happen if > virtualization is used (corosync aligns packets to a little bit less > then 1500 bytes, but bridge adds some bytes, so resulting packet may be > a little larger then 1500 bytes), so you can set netmtu in corosync.conf > to something like 1400. Right now sheepdog is using a single bridged nic. So setting netmtu may help. > - You can also try to set token value to little larger value (if you are > using 1000, use 2000, ...) My actual values are # How long before declaring a token lost (ms) token: 3000 # How many token retransmits before forming a new configuration token_retransmits_before_loss_const: 10 > But still keep in mind, that if there is HUGE network traffic, it can > happen that packets are lost (UDP). Ok, thank you very much. |