0
I have scripts that run on bootup on two NAS's (let's say they're QNAPs) to create bonds and have different ips for different vlans.
ie- NAS 1 (all NICs bonded to 1 via TLB) set bond0 -> 0.0.0.0 or flush bond 0 vlan 50 140.100.100.1 subnet 255.255.0.0
NAS 2 (all NICs bonded to 1 via TLB) set bond0 -> 0.0.0.0 or flush bond 0 vlan 50 140.100.100.2 subnet 255.255.0.0
My switch config looks something like ports 1 and 2 untagged for vlan 50 ports 5 and 6 tagged (NAS 1 connected to port 5, NAS 2 connected to port 6, computer connected to port 1 with proper network adapter configuration)
My problem is that when both NAS's are plugged in and I ping both at the same time from the computer, I get huge latency (>=32ms) on the pings and pretty much any traffic going to either unit. The second I unplug one of the NAS, the other NAS ping drops to <1ms and I no longer have bad latency. Initially I thought that it may be a IP collision issue so I look at ifconfig of the NAS's on startup - turns out it is running something after my autorun script to set the bond0s to a factory ip (150.140.100.100). So I manually set the bond ip to something arbitrary and different for both NAS's, turn off the /etc/init.d/dhcpd.sh script and try to ping 140.100.100.1 and 140.100.100.2 - it works for a few minutes then it goes back to timing out and having bad latency. This time however, bond0 of both NAS's have different ips - and that's where I am really confused. Thanks for any help! Also, MAC addresses on all the NICS are all different.
Summary - How do I fix latency issue between two NAS units connected to the same switch despite having different bond ips?
Thanks for your answer, I appreciate it. So I have mirroring bonding configurations on both NAS and switch (LACP). I always thought LACP provided fault tolerance so thats why I was testing by just plugging in 1/4 NICs from each NAS to test. However, plugging in all the NICS from the NAS into a 4x trunked LACP set of ports on the switch fixed the problem temporarily and I had latency again down the line. I monitor the port traffic and it turns out the two NAS are looping to each other tons of ARP request/DHCP packets for an unknown reason to me. – kenten – 2018-08-15T19:50:58.423
Because I am using a unique distro of linux on these QNAPs, they do not have ip netns so I have to use a hacky way to make multiple VLANs the same IP - by creating a bridge, and adding the VLANS on bond0 to the bridge, and then stopping it from acting like a switch. The reason for the bond is to allow multiple VLANs to access the bonded channel, and while it wont increase bandwidth, I want each access to be as close as the max bandwidth of a single port when all VLANs are acessing at the same time. Also the bonding on both NAS and switch creates failover (unplugged cable or switch down, etc) – kenten – 2018-08-15T19:55:08.897
i should mention that I am trying to achieve double redundancy on everything, two NAS, two switch trunked together, and all NICS on NAS bonded together with multiple VLAN access. The NAS needs to access multiple vlans with one IP. This problem with the two NAS looping tons of packets is really setting me back though, literally 300k packets a second and clogging the network. All my bridge does is accept traffic going to itself, and reject traffic from being broadcasted to the other vlans on the bridge - so that doesnt explain the huge amount of ARP / DHCP because those rules govern ip packets – kenten – 2018-08-15T20:01:39.897
Sorry, I don't see how multiple VLAN access helps with redundancy. And if you want redundancy, make sure to use a
backup
policy for bonding, not some policy that focuses on increase throughput. And funny network setups which cause network clogging are not going to help. Keep it simple. – dirkt – 2018-08-16T08:46:40.103sorry I misspoke, I meant to say bonded interfaces create redundancy via LACP or balance xor, etc - multiple vlan access is just a feature I need – kenten – 2018-08-16T17:33:04.730