交换机信息
本次配置的是 MQM9790-NS2X_Ax 交换机,他有 32 个 OSFP 端口,一个 OSFP 端口相当于 2 个 NDR 端口,所以共有 64 个 NDR 端口,每个 NDR 端口 400 Gbps,设置一分二就是给 NDR 端口设置一分二。
设置一分二
设置前需要先安装 MLNX_OFED 驱动,驱动下载地址:https://network.nvidia.com/products/infiniband-drivers/linux/mlnx_ofed/。
这个交换机和 HDR 的不一样,HDR 交换机设置一分二是配置 SPLIT_MODE
为 1 就行,NDR 交换机不仅要设置 SPLIT_MODE
为 1,让 OSFP 一分二,然后再对 NDR 端口单独设置 SPLIT_2X
。
这次使用的 IB 线是 NDR 一分四的线,800 Gbps 分 4 个 200 Gbps,分别是 1-1、1-2、2-1 和 2-2。可以把一个 800 Gbps OSFP 口分成 4 个 200 Gbps 口来用。
在未设置
SPLIT_MODE
为 1 时,一分四的线只有 1-1 号线可用,SPLIT_MODE
设置为 1 后,1-1 和 2-1 可用。将指定的端口设置
SPLIT_2X
后,1-1、1-2、2-1 和 2-2 四根线都可用。
查询交换机信息
[root@admin ~]# ibswitches
Switch : 0x0000000000000000 ports 129 "Quantum-2 Mellanox Technologies" base port 0 lid 2 lmc 0
重点是 lid 2
,这个是交换机的本地标识,后面会用到。
设置交换机为 SPLIT_MODE 模式
通过刚刚的 lid 2
来查询交换机的 SPLIT_MODE
模式。
[root@admin ~]# mlxconfig -d lid-2 q SPLIT_MODE
Device #1:
----------
Device type: Quantum2
Name: MQM9790-NS2X_Ax
Description: Mellanox Quantum 2 based NDR InfiniBand Switch; 64 NDR ports; 32 OSFP ports; non-blocking switching capacity of 51.2Tbps; 2 Power Supplies (AC); Standard depth; Unmanaged; airflow;
Device: lid-2
Configurations: Next Boot
SPLIT_MODE NO_SPLIT_SUPPORT(0)
[root@admin ~]# flint -d lid-2 swreset
-I- Sending reset command to device lid-2 ...
-I- Reset command accepted by the device.
修改 SPLIT_MODE
参数为 1:
[root@admin ~]# mlxconfig -d lid-2 s SPLIT_MODE=1
Device #1:
----------
Device type: Quantum2
Name: MQM9790-NS2X_Ax
Description: Mellanox Quantum 2 based NDR InfiniBand Switch; 64 NDR ports; 32 OSFP ports; non-blocking switching capacity of 51.2Tbps; 2 Power Supplies (AC); Standard depth; Unmanaged; airflow;
Device: lid-2
Configurations: Next Boot New
SPLIT_MODE NO_SPLIT_SUPPORT(0) SPLIT_2X(1)
重启交换机:
[root@admin ~]# flint -d lid-2 swreset
-I- Sending reset command to device lid-2 ...
-I- Reset command accepted by the device.
# 重启后查看模式修改是否成功
[root@admin ~]# mlxconfig -d lid-2 q SPLIT_MODE
Device #1:
----------
Device type: Quantum2
Name: MQM9790-NS2X_Ax
Description: Mellanox Quantum 2 based NDR InfiniBand Switch; 64 NDR ports; 32 OSFP ports; non-blocking switching capacity of 51.2Tbps; 2 Power Supplies (AC); Standard depth; Unmanaged; airflow;
Device: lid-2
Configurations: Next Boot
SPLIT_MODE SPLIT_2X(1)
重启后交换机变为 SPLIT_2X
模式,但是还没完,还需要针对 64 个 NDR 端口继续做设置,需要哪个口一分二就设置哪个口。
设置 NDR 端口为 SPLIT_2X 模式
查询端口模式
以 19 号端口为例:
[root@admin ~]# mlxconfig -d lid-2 q SPLIT_PORT[19]
Device #1:
----------
Device type: Quantum2
Name: MQM9790-NS2X_Ax
Description: Mellanox Quantum 2 based NDR InfiniBand Switch; 64 NDR ports; 32 OSFP ports; non-blocking switching capacity of 51.2Tbps; 2 Power Supplies (AC); Standard depth; Unmanaged; airflow;
Device: lid-2
Configurations: Next Boot
SPLIT_PORT[19] NO_SPLIT(0)
修改端口模式
[root@admin ~]# mlxconfig -d lid-2 s SPLIT_PORT[19]=1
Device #1:
----------
Device type: Quantum2
Name: MQM9790-NS2X_Ax
Description: Mellanox Quantum 2 based NDR InfiniBand Switch; 64 NDR ports; 32 OSFP ports; non-blocking switching capacity of 51.2Tbps; 2 Power Supplies (AC); Standard depth; Unmanaged; airflow;
Device: lid-2
Configurations: Next Boot New
SPLIT_PORT[19] NO_SPLIT(0) SPLIT_2X(1)
设置完仍然需要
flint -d lid-2 swreset
重启交换机。
端口支持批量设置和查询:
# 批量设置
[root@admin ~]# mlxconfig -d lid-2 s SPLIT_PORT[20..25]=1
Device #1:
----------
Device type: Quantum2
Name: MQM9790-NS2X_Ax
Description: Mellanox Quantum 2 based NDR InfiniBand Switch; 64 NDR ports; 32 OSFP ports; non-blocking switching capacity of 51.2Tbps; 2 Power Supplies (AC); Standard depth; Unmanaged; airflow;
Device: lid-2
Configurations: Next Boot New
SPLIT_PORT[20] NO_SPLIT(0) SPLIT_2X(1)
SPLIT_PORT[21] NO_SPLIT(0) SPLIT_2X(1)
SPLIT_PORT[22] NO_SPLIT(0) SPLIT_2X(1)
SPLIT_PORT[23] NO_SPLIT(0) SPLIT_2X(1)
SPLIT_PORT[24] NO_SPLIT(0) SPLIT_2X(1)
SPLIT_PORT[25] NO_SPLIT(0) SPLIT_2X(1)
Apply new Configuration? (y/n) [n] : y
Applying... Done!
-I- Please reboot machine to load new configurations.
# 批量查询
[root@admin ~]# mlxconfig -d lid-2 q SPLIT_PORT[20..25]
Device #1:
----------
Device type: Quantum2
Name: MQM9790-NS2X_Ax
Description: Mellanox Quantum 2 based NDR InfiniBand Switch; 64 NDR ports; 32 OSFP ports; non-blocking switching capacity of 51.2Tbps; 2 Power Supplies (AC); Standard depth; Unmanaged; airflow;
Device: lid-2
Configurations: Next Boot
SPLIT_PORT[20] SPLIT_2X(1)
SPLIT_PORT[21] SPLIT_2X(1)
SPLIT_PORT[22] SPLIT_2X(1)
SPLIT_PORT[23] SPLIT_2X(1)
SPLIT_PORT[24] SPLIT_2X(1)
SPLIT_PORT[25] SPLIT_2X(1)
查看设置一分二之后的连接信息
[root@admin ~]# iblinkinfo
CA: comput1 HCA-1:
0x0000000000000000 5 1[ ] ==( 2X 106.25 Gbps Active/ LinkUp)==> 2 22[ ] "Quantum-2 Mellanox Technologies" (Could be 4X )
CA: Mellanox Technologies Aggregation Node:
0x0000000000000000 3 1[ ] ==( 4X 106.25 Gbps Active/ LinkUp)==> 2 65[ ] "Quantum-2 Mellanox Technologies" ( )
Switch: 0x0000000000000000 Quantum-2 Mellanox Technologies:
2 1[ ] ==( Down/ Polling)==> [ ] "" ( )
2 2[ ] ==( Down/ Polling)==> [ ] "" ( )
...output omitted...
2 20[ ] ==( Down/ Polling)==> [ ] "" ( )
2 21[ ] ==( 2X 106.25 Gbps Active/ LinkUp)==> 1 1[ ] "admin HCA-1" (Could be 4X )
2 22[ ] ==( 2X 106.25 Gbps Active/ LinkUp)==> 5 1[ ] "comput1 HCA-1" (Could be 4X )
2 23[ ] ==( Down/ Polling)==> [ ] "" ( )
...output omitted...
2 63[ ] ==( Down/ Polling)==> [ ] "" ( )
2 64[ ] ==( Down/ Polling)==> [ ] "" ( )
2 65[ ] ==( 4X 106.25 Gbps Active/ LinkUp)==> 3 1[ ] "Mellanox Technologies Aggregation Node" ( )
CA: admin HCA-1:
0x0000000000000000 1 1[ ] ==( 2X 106.25 Gbps Active/ LinkUp)==> 2 21[ ] "Quantum-2 Mellanox Technologies" (Could be 4X )