Core routers
We are using 2 hosts for routing purposes: rc1v.itim.vn and rc2v.itim.vn
Network Interfaces
Both hosts has 2 integrated 1G ports and 4x1G ports from installed Intel ET2 Quad port NIC
All 6 ports aggregated into LACP etherchannel (bond0 inteface on server):
auto bond0
iface bond0 inet manual
bond-slaves eth0 eth1 eth2 eth3 eth4 eth5
bond-mode 4
bond-miimon 100
bond-downdelay 200
bond-updelay 200
bond_lacp_rate fast
bond_xmit_hash_policy layer3+4
SMP Affinity
IRQs for TxRx? splitted per CPU core by script in itim-irq deb package.
RX/TX Queue tuning
NIC offloads
Failover
- rc1v is working as primary (VRRP vmac-based) default gateway for:
- vlan 100 – Production search and services hosts (management, main updates)
- vlan 101 – Vlan for crawlers and indexers (on 18.06.2013 – all crawlers and indexers still in dev vlan)
- vlan 700 – Networking devices
- vlan 778 – IPMI and infrastructure
- vlan 1050 – Development hosts
- If rc1v is down, keepalived on rc2v must change state to MASTER. What does mean down state:
- vrrp_sync_group RC_VDC is in Fault state
- Any one instance of vrrp_sync_group RC_VDC in fault state
- Any interface from list below is down in any instance:
bond0 bond0.100 bond0.101 bond0.777 bond0.778 bond0.998 bond0.1000 bond0.1050
- What does mean MASTER – all interfaces from the list below is up:
vrrp.10.100 10.10.0.1/23 vrrp.11.101 10.10.64.1/23 vrrp.12.777 10.100.0.1/24 vrrp.13.778 10.101.0.1/22 vrrp.14.1050 10.3.0.1/22
keepalived version: 1.2.7
Keepalived Scripts
Routing
We are using quagga package ( 0.99.17-2+squeeze3 BGP/OSPF/RIP routing daemon ) from Debian Squeeze repo for OSPF.
OSPF and VRRP – based failover schema: