Lustre利用LNet支持多种协议之间的互通,比如InfiniBand(IB)、Intel Omni-Path(OPA)与以太网等,并且支持多路由冗余等,本文主要介绍IB与OPA之间的互通。
|
注意:服务端与客户端的网络类型要不同,如分别为o2ib1和o2ib2,LNet端与服务端和客户端互联的类型要分别一一对应。
systemctl disable firewalld
systemctl stop firewalld
ip route add 10.11.0.0/16 via 10.10.100.11
上述设置系统重启后会失效,可以添加到启动脚本/etc/rc.local中设置成设成永久有效:
echo "ip route add 10.11.0.0/16 via 10.10.100.11">>/etc/rc.local
default via 192.168.100.254 dev eno1 proto static metric 100 10.10.0.0/16 dev ib0 proto kernel scope link src 10.10.100.1 10.11.0.0/16 via 10.10.100.11 dev ib0 169.254.0.0/16 dev ib0 scope link metric 1007 192.168.0.0/16 dev eno1 proto kernel scope link src 192.168.100.1 metric 100
#选项 lnet networks="本机IB网络类型(网卡)" routes="客户端OPA网络类型 LNet节点对应本地服务端IB的IP@网络类型"
options lnet networks="o2ib1(ib0)" routes="o2ib2 10.10.100.[11-12]@o2ib1"
lustre_rmmod
modprobe -v lustre
mkfs.lustre --fsname=lxfs --mgs --mdt --index=0 --servicenode=10.10.100.1@o2ib1 --servicenode=10.10.100.2@o2ib1 --reformat /dev/mapper/mpathb
mkfs.lustre --fsname=lxfs --mgsnode=10.10.100.1@o2ib1 --mgsnode=10.10.100.2@o2ib1 --servicenode=10.10.100.1@o2ib1 --servicenode=10.10.100.2@o2ib1 --ost --reformat --index=1 /dev/mapper/mpathc
mkfs.lustre --fsname=lxfs --mgsnode=10.10.100.1@o2ib1 --mgsnode=10.10.100.2@o2ib1 --servicenode=10.10.100.1@o2ib1 --servicenode=10.10.100.2@o2ib1 --ost --reformat --index=2 /dev/mapper/mpathd
10.10.100.1@o2ib1
net: - net type: lo local NI(s): - nid: 0@lo status: up - net type: o2ib1 local NI(s): - nid: 10.10.100.1@o2ib1 status: up interfaces: 0: ib0
route: - net: o2ib2 gateway: 10.10.100.11@o2ib1 - net: o2ib2 gateway: 10.10.100.12@o2ib1
default via 192.168.100.254 dev ens5 proto static metric 100 10.10.0.0/16 dev ib0 proto kernel scope link src 10.10.100.11 metric 150 10.11.0.0/16 dev ib1 proto kernel scope link src 10.11.100.11 metric 150 192.168.0.0/16 dev ens5 proto kernel scope link src 192.168.100.11 metric 100
echo 1 > /proc/sys/net/ipv4/conf/ib0/forwarding
echo 1 > /proc/sys/net/ipv4/conf/ib1/forwarding
上述设置系统重启后会失效,编辑/usr/lib/sysctl.d/00-system.conf设成永久有效(仅对ib0和ib1的IPv4进行转发):
net.ipv4.conf.ib0.forwarding=1
net.ipv4.conf.ib1.forwarding=1
#选项 lnet networks="本机IB网络类型(IB网卡),本机OPA网络类型(OPA网卡)" forwarding="启用"
options lnet networks="o2ib1(ib0),o2ib2(ib1)" forwarding="enabled"
lustre_rmmod
modprobe -v lustre
10.10.100.11@o2ib1 10.11.100.11@o2ib2
net: - net type: lo local NI(s): - nid: 0@lo status: up - net type: o2ib1 local NI(s): - nid: 10.10.100.11@o2ib1 status: up interfaces: 0: ib0 - net type: o2ib2 local NI(s): - nid: 10.11.100.11@o2ib2 status: up interfaces: 0: ib1
ip route add 10.10.0.0/16 via 10.11.100.11
上述设置系统重启后会失效,可以采用以下两种方法之一设置成设成永久有效:
echo "ip route add 10.11.0.0/16 via 10.10.100.11">>/etc/rc.local
10.10.0.0/16 via 10.11.100.11 dev ib0
default via 192.168.100.254 dev enp2s0f0 proto static metric 100 10.10.0.0/16 via 10.11.100.11 dev ib0 10.11.0.0/16 dev ib0 proto kernel scope link src 10.11.135.1 169.254.0.0/16 dev ib0 scope link metric 1004 192.168.0.0/16 dev enp2s0f0 proto kernel scope link src 192.168.135.1 metric 100
#选项 lnet networks="本机客户端OPA网络类型(网卡)" routes="服务端IB网络类型 LNet节点对应客户端OPA的IP@网络类型"
options lnet networks="o2ib2(ib0)" routes="o2ib1 10.11.100.[11-12]@o2ib2"
lustre_rmmod
modprobe -v lustre
10.11.135.1@o2ib2
net: - net type: lo local NI(s): - nid: 0@lo status: up - net type: o2ib2 local NI(s): - nid: 10.11.135.1@o2ib2 status: up interfaces: 0: ib0
route: - net: o2ib1 gateway: 10.11.100.11@o2ib2 - net: o2ib1 gateway: 10.11.100.12@o2ib2
mount -v -t lustre 10.10.100.1@o2ib1:/lxfs /home
Filesystem 1K-blocks Used Available Use% Mounted on 10.10.100.1@o2ib1:/lxfs 7341068688 143568 6970159092 1% /home
traceroute to 10.11.135.1 (10.11.135.1), 30 hops max, 60 byte packets 1 10.10.100.11 (10.10.100.11) 0.097 ms 0.064 ms 0.048 ms 2 10.11.135.1 (10.11.135.1) 0.146 ms 0.135 ms 0.126 ms
traceroute to 10.10.100.1 (10.10.100.1), 30 hops max, 60 byte packets 1 10.11.100.11 (10.11.100.11) 0.105 ms 0.082 ms 0.066 ms 2 bio1 (10.10.100.1) 0.149 ms 0.147 ms 0.145 ms
12345-0@lo 12345-10.10.100.11@o2ib1 12345-10.11.100.11@o2ib2
12345-0@lo 12345-10.11.135.1@o2ib2
12345-0@lo 12345-10.10.100.1@o2ib1
参考文档:Intel ® Omni-Path Storage Router(http://www.intel.com/content/dam/support/us/en/documents/network-and-i-o/fabric-products/Intel_OP_Storage_Router_DG_H99668_v2_0.pdf)