Hi All,
Appear to have found a bug in XCP where bonded interfaces with jumbo
frames (mtu 9000) are not being brought up properly on bootup. The
bonded interface is setup with mtu 9000, but the underlying interfaces
are left at mtu 1500.
I've got my host configured as such:
eth0 + eth1 = Bonded together with mtu 1500 for internet access
eth2 + eth3 = Bonded together with mtu 9000 for mgmt / storage
Management IP is configured on the underlying eth2+3 bond
Storage IPs are configured on vlan interfaces on top of the eth2+3 bond
This all works awesome immediately after configuration, mgmt traffic
leaves the bond untagged and storage traffic works as expected.
On reboot of the host however, both eth2 + eth3 are brought up with only
mtu 1500:
eth2 Link encap:Ethernet HWaddr A0:36:9F:0E:D8:54
UP BROADCAST RUNNING PROMISC MULTICAST MTU:1500 Metric:1
eth3 Link encap:Ethernet HWaddr A0:36:9F:0E:D8:56
UP BROADCAST RUNNING PROMISC MULTICAST MTU:1500 Metric:1
Even though the management interface has mtu 9000 set:
xapi2 Link encap:Ethernet HWaddr A0:36:9F:0E:D8:54
inet addr:10.0.16.13 Bcast:10.0.16.255 Mask:255.255.255.0
UP BROADCAST RUNNING PROMISC MULTICAST MTU:9000 Metric:1
This results in network apparently working - icmp works, you can ssh
somewhere but as soon as a large burst of traffic happens then the tcp
connection stalls as it gets fragmented weirdly. Another host which has
never been rebooted since configuration (via xencenter) that it has
eth2/eth3 properly set with mtu 9000.
Manually doing an "ifconfig ethX mtu 9000" for eth2 + eth3, followed by
an xe-toolstack-restart results in a working host again.
So, pretty sure this is a bug =)
- Nathan
_______________________________________________
Xen-api mailing list
Xen-api@lists.xen.org
http://lists.xen.org/cgi-bin/mailman/listinfo/xen-api
