Skip to content

H3C RoCE 交换机配置 ​

H3C RoCE 交换机管理、VLAN、端口拆分、PFC/ECN 及 MXRDMA 配置实例。

返回交换机配置索引 | 下一篇:锐捷 RoCE 交换机配置


H3C RoCE 技术资料

常用的基本操作 ​

分配网络管理口和账户 ​

进入交换机命令行

使用命令

sys

进入特权模式

使用命令

sysname MUXI

更改设备名为:MUXI

分配网络管理口和账户

使用命令

bash
ssh server enable

开启ssh登录

使用命令

bash
ssh user

创建 SSH 用户,并指定 SSH 用户的服务类型和认证方式

undo ssh user 命令用来删除 SSH 用户

分配网络管理口和账户

配置密码

分配网络管理口和账户

分配网络管理口和账户

配置管理口IP地址

分配网络管理口和账户

退出保存配置

exit

write

在别的网段的机器上面ssh登录,验证配置是否成功

分配网络管理口和账户

查看信息 ​

进入特权模式

sys

端口连接状态

display interface brief

查看当前配置

display current-configuration

端口拆分和恢复 ​

400G 拆分成 200G

interface 400GE1/0/1 //单个拆分

interface range 400GE1/0/1 to 400GE1/0/35 //范围拆分(只支持拆分,不支持范围恢复)

using twohundredgige

400G 拆分成 200G后恢复400G

using fourhundredgige //恢复只能一个一个操作,或者自己写脚本批量

拆分后如果网口没有UP,尝试下面命令使能端口

undo shutdown

端口拆分和恢复

丢包查看 ​

先确认要查看的网卡对应网口,服务器侧通过"lldpcli show neigh"命令来查看网卡连接的交换机端口

丢包查看

查看网络是否丢包和ECN触发次数1

display packet-drop interface HGE1/1/16

丢包查看

重置丢包信息:

reset packet-drop interface HGE1/1/16

reset counters interface HGE1/1/16

查看网络PFC触发次数

display priority-flow-control interface

vlan配置 ​

新增vlan

vlan配置

vlan配置

上图ip address后面填写的是网关地址,也就是yaml文件中的via字段后面跟的ip地址

下图是某次配置的截图:

vlan配置

ip address执行完毕后,需要执行undo shutdown命令

display this的结果如图:

vlan配置

添加端口到vlan

vlan配置

端口使用lldpcli show neigh命令查看,下图框起来的就是port后需要跟的端口号

vlan配置

根据该图,执行如下命令即可将此两个端口加到vlan中,如下图

vlan配置

添加完毕后,使用下图命令查看vlan 2的内容

vlan配置

最后sudo netplan apply或者在最开始执行这个命令也行

show_gids不连续 ​

执行完4.2.2节后,show_gids发现gid不连续?是因为在sudo netplan apply的时候,某个网卡正在被使用,某个GID正在被占用,所以你在执行sudo netplan apply的时候,在原先GID的基础上递增。

如何解决?只需要关闭掉正在使用这些网卡的应用,再次执行suod netplan apply即可

100Gbps交换机的推荐配置 ​

假设使用优先级 5 来承载分布式存储的 RoCEv2 流量。在交换机里,各个面板端 口需要针对优先级 5 使能 PFC 流控,命令行如下。

bash
sys //进入特权模式

#全局配置
buffer egress cell queue 5 shared ratio 100 //共享缓冲区 出口队列五可以使用率100%

buffer egress cell queue 6 shared ratio 100
buffer apply

qos wred queue table QOS-EGRESS-100G-PORT
queue 5 low-limit 1000 high-limit 2000 discard-probability 20
queue 5 weighting-constant 0
queue 5 ecn

priority-flow-control poolID 0 headroom 130000  //配置为允许的最大值
priority-flow-control deadlock cos 5 interval 10
priority-flow-control deadlock precision high

#配置端口HundredGigE1/0/1到interface HundredGigE1/0/35
interface range 100GE1/0/1 to 100GE1/0/35
qos wred apply QOS-EGRESS-100G-PORT
qos trust dscp
priority-flow-control enable
priority-flow-control no-drop dot1p 5
priority-flow-control dot1p 5 headroom 491
priority-flow-control dot1p 5 ingress-buffer dynamic 5
priority-flow-control dot1p 5 ingress-threshold-offset 12
priority-flow-control deadlock enable

#保存配置

exit

save

400Gbps交换机的推荐配置 ​

假设使用优先级 5 来承载分布式存储的 RoCEv2 流量。在交换机里,各个面板端口需要针对优先级 5 使能 PFC 流控,命令行如下。

bash
sys //进入特权模式

buffer egress cell queue 5 shared ratio 100 //共享缓冲区 出口队列五可以使用率100%

buffer egress cell queue 6 shared ratio 100
buffer apply

qos wred queue table QOS-EGRESS-400G-PORT

queue 5 drop-level 0 low-limit 2100 high-limit 5000 discard-probability 20

queue 5 drop-level 1 low-limit 2100 high-limit 5000 discard-probability 20

queue 5 drop-level 2 low-limit 2100 high-limit 5000 discard-probability 20

queue 5 weighting-constant 0

queue 5 ecn

priority-flow-control poolid 0 headroom 64000

priority-flow-control deadlock cos 5 interval 10

priority-flow-control deadlock precision high

#配置端口FourHundredGigE1/0/1到interface FourHundredGigE1/0/35,通过dis interface b 查看可用端口

interface range 400GE1/0/1 to 400GE1/0/35

qos trust dscp

qos wred apply QOS-EGRESS-400G-PORT

priority-flow-control enable

priority-flow-control no-drop dot1p 5 pause-threshold ratio 5 pause-threshold-offset 12 reserved-buffer 16

priority-flow-control deadlock enable

exit

save

MXRDMA网卡交换机配置实例 ​

N300和C600机器连接交换机的方式不同

N300是MXRDMA网卡连接200G端口的交换机H3C-S9825,128口,

C600是将两个MXRDMA网卡通过转接器合并到400G的线接入400G双端口(每个端口200G)的交换机H3C-S9857。

N300 MXRDMA 交换机配置 ​

目前针对型号H3C-S9825-128B。

交换机上创建vlan 2:

  • 根据服务器上IB网卡个数为vlan 2分配IP,如服务器上有16个网卡,则分配16个IP地址。

  • 将网卡连接的16个交换机端口加入到vlan 2

# 进入vlan 2配置
interface vlan 2

# 进入特权模式
system-view
# 创建vlan 2
vlan 2
#端口接入vlan 2
port 200GE1/0/1
port 200GE1/0/2
port 200GE1/0/3
port 200GE1/0/4
port 200GE1/0/5
port 200GE1/0/6
port 200GE1/0/7
port 200GE1/0/8
port 200GE1/0/9
port 200GE1/0/10
port 200GE1/0/11
port 200GE1/0/12
port 200GE1/0/13
port 200GE1/0/14
port 200GE1/0/15
port 200GE1/0/16

#配置vlan的网关
interface vlan 2
undo ipv6 forwarding
ip address 11.22.100.1 255.255.255.0
ip address 11.22.101.1 255.255.255.0 sub
ip address 11.22.102.1 255.255.255.0 sub
ip address 11.22.103.1 255.255.255.0 sub
ip address 11.22.104.1 255.255.255.0 sub
ip address 11.22.105.1 255.255.255.0 sub
ip address 11.22.106.1 255.255.255.0 sub
ip address 11.22.107.1 255.255.255.0 sub
ip address 11.22.108.1 255.255.255.0 sub
ip address 11.22.109.1 255.255.255.0 sub
ip address 11.22.100.1 255.255.255.0 sub
ip address 11.22.111.1 255.255.255.0 sub
ip address 11.22.112.1 255.255.255.0 sub
ip address 11.22.113.1 255.255.255.0 sub
ip address 11.22.114.1 255.255.255.0 sub
ip address 11.22.115.1 255.255.255.0 sub
# 退出vlan配置
exit
# 保存vlan配置
write

交换机上连接服务器网卡的端口配置,所有的端口配置QoS。

# 进入特权模式
system-view
# 配置端口1-16,注意根据实际情况配置实际的端口号
interface range TwoHundredGigE 1/0/1 to TwoHundredGigE 1/0/16
# 进行配置
port link-mode bridge
port access vlan 2
link-delay down 0
link-delay up 2
# 为队列5配置PFC
priority-flow-control enable
priority-flow-control no-drop dot1p 5 pause-threshold ratio 12 headroom 2000
priority-flow-control deadlock enable
flow-interval 5
# 配置ecn dscp
qos trust dscp
qos wfq byte-count
qos wfq ef group 1 byte-count 96
qos wfq cs6 group sp
qos wfq cs7 group sp
qos wred queue 5 drop-level 0 low-limit 3500 high-limit 5000 discard-probability 40
qos wred queue 5 drop-level 1 low-limit 3500 high-limit 5000 discard-probability 40
qos wred queue 5 drop-level 2 low-limit 3500 high-limit 5000 discard-probability 40
qos wred queue 5 ecn
qos wred queue 5 weighting-constant 0

C600 MXRDMA 交换机配置 ​

交换机上创建vlan 2:

  • 根据服务器上IB网卡个数为vlan 2分配IP,如服务器上有16个网卡,则分配16个IP地址。

  • 将网卡连接的16个交换机端口加入到vlan 2

H3C-S9827。

port link-mode bridge
port access vlan 2
priority-flow-control enable
priority-flow-control no-drop dot1p 5 pause-threshold ratio 5
priority-flow-control deadlock enable
qos trust dscp
qos wfq byte-count
qos wfq ef group 1 byte-count 96
qos wfq cs6 group sp
qos wfq cs7 group sp
qos wred queue 5 drop-level 0 low-1imit 6000 high-1imit 12000 discard-probability 40
qos wred queue 5 drop-level 1 low-limit 6000 high-limit 12000 discard-probability 40
qos wred queue 5 drop-level 2 low-limit 6000 high-limit 12000 discard-probability 40
qos wred queue5 ecn
qos wred queue 5 weighting-constant 0

H3C-S9827-128DH

port link-mode bridge
port access vlan 2
priority-flow-control enable
priority-flow-control no-drop dot1p 5 pause-threshold ratio 6 headroom 1200
priority-flow-control deadlock enable
qos trust dscp
qos wfq byte-count
qos wfq ef group 1 byte-count 96
qos wfq cs6 group sp
qos wfq cs7 group sp
qos wred queue 5 drop-level 0 low-1imit 3500 high-1imit 8000 discard-probability 40
qos wred queue 5 drop-level 1 low-limit 3500 high-limit 8000 discard-probability 40
qos wred queue 5 drop-level 2 low-limit 3500 high-limit 8000 discard-probability 40
qos wred queue5 ecn
qos wred queue 5 weighting-constant 0

C600 MXRDMA 交换机配置

星融元CX864-E

ssh登录交换机

C600 MXRDMA 交换机配置

普通视图只能查看信息,配置需要config视图(config视图查看信息需要do show xxx)

C600 MXRDMA 交换机配置

vlan配置

!
vlan range 100-300
!
interface vlan 100
ip address 11.22.100.1/24
ip address 11.22.101.1/24
ip address 11.22.102.1/24
ip address 11.22.103.1/24
ip address 11.22.104.1/24
ip address 11.22.105.1/24
ip address 11.22.106.1/24
ip address 11.22.107.1/24
ip address 11.22.108.1/24
ip address 11.22.109.1/24
ip address 11.22.110.1/24
ip address 11.22.111.1/24
ip address 11.22.112.1/24
ip address 11.22.113.1/24
ip address 11.22.114.1/24
ip address 11.22.115.1/24
exit
!

接口配置

!
interface ethernet 0/0
service-policy roce_lossless_40m_low_balance_200g
breakout 4x200G[100G]
fec rs
mtu 9216
speed 200000
switchport access vlan 100
exit
!

接口绑定RoCE

!
qos roce roce_lossless_40m_low_balance
bind interface range 0/1:1-0/64:4
exit
!

RoCE配置(星融元内置模板,请谨慎改动)

!
buffer-profile egress_lossless_profile
mode lossless static 22798336 size 0
exit
!
buffer-profile egress_lossy_profile
mode lossy dynamic 2 size 1518
exit
!
buffer-profile ingress_lossless_profile
mode lossless dynamic 1 size 1518 xon-offset 9408 xoff 135520
exit
!
buffer-profile ingress_lossy_profile
mode lossy static 22798336 size 0
exit
!
buffer-profile pg_lossless_10000_40m_profile
mode lossless dynamic 1 size 1518 xon-offset 9408 xon 0 xoff 16928
exit
!
buffer-profile pg_lossless_25000_40m_profile
mode lossless dynamic 1 size 1518 xon-offset 9408 xon 0 xoff 16928
exit
!
buffer-profile pg_lossless_50000_40m_profile
mode lossless dynamic 1 size 1518 xon-offset 9408 xon 0 xoff 23392
exit
!
buffer-profile pg_lossless_100000_40m_profile
mode lossless dynamic 1 size 1518 xon-offset 9408 xon 0 xoff 38816
exit
!
buffer-profile pg_lossless_200000_40m_profile
mode lossless dynamic 1 size 1518 xon-offset 9408 xon 0 xoff 71904
exit
!
buffer-profile pg_lossless_400000_40m_profile
mode lossless dynamic 1 size 1518 xon-offset 9408 xon 0 xoff 135520
exit
!
buffer-profile pg_lossless_800000_40m_profile
mode lossless dynamic 1 size 1518 xon-offset 9408 xon 0 xoff 205520
exit
!
buffer-profile roce_lossless_40m_low_balance_200g_profile
mode lossless dynamic 1 size 1518 xon-offset 9408 xoff 71904
exit
!
wred default_drop
mode drop gmin 1048576 gmax 2097152 gprobability 10 ymin 1048576 ymax 2097152 yprobability 10 rmin 1048576 rmax 2097152 rprobability 10
exit
!
wred default_ecn
mode ecn gmin 1048576 gmax 2097152 gprobability 10 ymin 1048576 ymax 2097152 yprobability 10 rmin 1048576 rmax 2097152 rprobability 10
exit
!
wred roce_lossless_40m_low_balance_200g_ecn
mode ecn gmin 1768622 gmax 8926720 gprobability 50
exit
!
priority-flow-control enable 1 5
!
class-map roce_lossless_40m_low_balance_200g_class_map
match cos 1 5
exit
!
policy-map roce_lossless_40m_low_balance_200g
class roce_lossless_40m_low_balance_200g_class_map
priority-group-buffer roce_lossless_40m_low_balance_200g_profile
wred roce_lossless_40m_low_balance_200g_ecn
exit
queue-scheduler priority queue 6
queue-scheduler priority queue 7
set cos dscp diffserv roce_lossless_diffserv_map
exit
!

返回交换机配置索引 | 下一篇:锐捷 RoCE 交换机配置