近期在做集群网络的升级方案,涉及到网络设备有交换机、线缆、网卡。过程中测试了不少网络厂商的设备,这里主要记录下在测试Chelsio网卡时的一些问题。
测试内容
lustre on iwarp
网卡带宽和延迟这里不做过多介绍,这个测试起来比较简单,直接跑下iperf和ib_write测试工具即可;这里主要记录下如何在lustre上启用iwarp。
直接上图:
存储硬件设备使用DDN/SFA7700x作为lustre的后端盘阵,共60块7.2K NL-SAS硬盘。存储与IO节点之间使用IB互联,IO节点对外使用Chelsio T580-LP-CR 40Gb网卡,客户端有2个,一个是40Gb网卡,一个是10Gb网卡。
虽然Mellonax和Chelsio网卡驱动包里都包含了ofed驱动,但如果把两张网卡放在一台机器上使用并加载iwarp功能,ofed驱动会不兼容,这里mellonax厂商可能对其驱动包里的ofed做了修改,阉割了对iwarp的支持。所以我们第一步就是安装开源ofed。
[root@hpc-242-243 ~]# cd OFED-3.18-1
[root@hpc-242-243 OFED-3.18-1]# ls
BUILD_ID docs install.pl LICENSE ofed.conf.save README.txt RPMS SOURCES SPECS SRPMS uninstall.sh
[root@hpc-242-243 OFED-3.18-1]# ./install.pl
执行install.pl脚本,根据提示安装即可。安装过程中可能会报错,有些依赖软件需要手动安装
[root@hpc-242-243 ~]#yum install openldap-devel apr-util-devel mod_ssl byacc.x86_64 libtool kernel-devel redhat-rpm-config rpm-build
我的环境里缺的是这些组件。
[root@hpc-242-244 ~]# cd ChelsioUwire-2.12.0.2
[root@hpc-242-244 ChelsioUwire-2.12.0.2]# ls
build dialog.py docs install.log libs Makefile rpmbuild sample_machinefile specs support Uboot debrules dialog.pyc EULA install.py lustre OFED RPM-Manager scripts src tools uninstall.log [root@hpc-242-244 ChelsioUwire-2.12.0.2]# ./install.py -c nic ------------------------------- Chelsio Unified Installer v2.0 ------------------------------- Error : The kernel version 2.6.32-504.3.3.el6_lustre.es131.x86_64 is not supported. Refer to README for supported kernel versions. To compile drivers for an updated kernel, press (Y). To exit, press (N).
Input(Y/N):y ----------------------------------------------------------------------------
| Choice | Kernel Version | Distro |
| Press 1 | 2.6.32-279.el6 | Red Hat Enterprise Linux Server release 6.3 |
| Press 2 | 2.6.32-358.el6 | Red Hat Enterprise Linux Server release 6.4 |
| Press 3 | 2.6.32-431.el6 | Red Hat Enterprise Linux Server release 6.5 |
| Press 4 | 2.6.32-504.el6 | Red Hat Enterprise Linux Server release 6.6 |
| Press 5 | 3.10.0-123.el7 | Red Hat Enterprise Linux Server release 7 |
| Press 6 | 3.10.0-229.el7 | Red Hat Enterprise Linux Server release 7.1 |
| Press 7 | 3.10.0-327.el7 | Red Hat Enterprise Linux Server release 7.2 |
| Press 8 | 2.6.32.12-0.7 | SUSE Linux Enterprise Server 11 SP1 |
| Press 9 | 3.0.13-0.27 | SUSE Linux Enterprise Server 11 SP2 |
| Press 10 | 3.0.76-0.11 | SUSE Linux Enterprise Server 11 SP3 |
| Press 11 | 3.0.101-63 | SUSE Linux Enterprise Server 11 SP4 |
| Press 12 | 3.12.28-4 | SUSE Linux Enterprise Server 12 |
| Press 13 | 3.12.49-11 | SUSE Linux Enterprise Server 12 SP1 |
| Press 14 | 3.13.0-32 | Ubuntu 14.04.1 LTS |
| Press 15 | 3.16.0-30 | Ubuntu 14.04.2 LTS |
| Press 16 | 3.19.0-25 | Ubuntu 14.04.3 LTS |
| Press 17 | 3.4 | Linux kernel Release |
| Press 18 | 3.6 | Linux kernel Release |
| Press 19 | 3.7 | Linux kernel Release |
| Press 20 | 3.8 | Linux kernel Release |
| Press 21 | 3.9 | Linux kernel Release |
| Press 22 | 3.10 | Linux kernel Release |
| Press 23 | 3.11 | Linux kernel Release |
| Press 24 | 3.12 | Linux kernel Release |
| Press 25 | 3.13 | Linux kernel Release |
| Press 26 | 3.14 | Linux kernel Release |
| Press 27 | 3.15 | Linux kernel Release |
| Press 28 | 3.16 | Linux kernel Release |
| Press 29 | 3.17 | Linux kernel Release |
| Press 30 | 3.18 | Linux kernel Release |
| Press 31 | 4.1 | Linux kernel Release |
| Press 32 | exit | | ----------------------------------------------------------------------------
Input:4 -------------------------------
Chelsio Unified Installer v2.0 -------------------------------
-------------------------------------------------------------------------------------------- | Choice | T4/T5 Configuration | Supported Protocols/Drivers | --------------------------------------------------------------------------------------------
| Press 1 | Unified Wire | all the Chelsio drivers | --------------------------------------------------------------------------------------------
| Press 2 | Low Latency Networks | NIC;TOE;RDMA;WD | --------------------------------------------------------------------------------------------
| Press 3 | High Capacity RDMA | NIC;TOE;RDMA | --------------------------------------------------------------------------------------------
| Press 4 | RDMA Performance | NIC;TOE;RDMA | --------------------------------------------------------------------------------------------
| Press 5 | High Capacity TOE | NIC;TOE | --------------------------------------------------------------------------------------------
| Press 6 | iSCSI Performance | NIC;TOE;iSCSI | --------------------------------------------------------------------------------------------
| Press 7 | UDP Segmentation Offload & Pacing | UDP segmenation offload capable NIC;TOE | --------------------------------------------------------------------------------------------
| Press 8 | T5 Wire Direct Latency | NIC;TOE;RDMA;WD | --------------------------------------------------------------------------------------------
| Press 9 | T5 High Capacity WD | NIC;RDMA;WD | --------------------------------------------------------------------------------------------
| Press 10 | T5 Hash Filter Configuration | NIC | --------------------------------------------------------------------------------------------
| Press 11 | EXIT | | --------------------------------------------------------------------------------------------
Input:1
------------------------------- Chelsio Unified Installer v2.0 -------------------------------
---------------------------------------------------------------------- | Choice | OFED Configuration | Details | ----------------------------------------------------------------------
| Press 1 | Install OFED | Compiles and Installs OFED 3.18-1 | ----------------------------------------------------------------------
| Press 2 | Skip OFED | Skips Ofed install | ----------------------------------------------------------------------
| Press 3 | EXIT | | ----------------------------------------------------------------------
Input:2 -------------------------------
Chelsio Unified Installer v2.0 -------------------------------
*****************
Installer Summary
*****************
Targets provided : nic tools
Checking Build : Done
Checking Targets : Done
Checking Tunnables : Done
Checking Dependencies : Done
Building and Installing :
正常的话,这样结束就可以了。不过安装过程中可能还是会报一些错误,不过日志里都有记录,根据日志提示一个一个解决即可。刚才只是安装了网卡的驱动,还没有安装iwarp相关软件包,需要再执行一下./install.py -c iwarp。安装方法和上面一样。
安装完成后重启一下系统,然后加载相关模块即可:
[root@hpc-242-243 ~]# modprobe cxgb4
[root@hpc-242-243 ~]# modprobe iw_cxgb4
[root@hpc-242-243 ~]# ibv_devices
device node GUID
------ ----------------
cxgb4_0 0007432dd4700000
mlx4_0 14dda9d424c80000
[root@hpc-242-243 ~]# ibv_devinfo
hca_id: cxgb4_0
transport: iWARP (1)
fw_ver: 0.271.7168
node_guid: 0007:432d:d470:0000
sys_image_guid: 0007:432d:d470:0000
vendor_id: 0x1425
vendor_part_id: 21520
hw_ver: 0x0
board_id: 1425.5410
phys_port_cnt: 2
port: 1
state: PORT_DOWN (1)
max_mtu: 4096 (5)
active_mtu: 1024 (3)
sm_lid: 0
port_lid: 0
port_lmc: 0x00
link_layer: Ethernet
port: 2
state: PORT_ACTIVE (4)
max_mtu: 4096 (5)
active_mtu: 1024 (3)
sm_lid: 0
port_lid: 0
port_lmc: 0x00
link_layer: Ethernet
hca_id: mlx4_0
transport: InfiniBand (0)
fw_ver: 2.32.5100
node_guid: 14dd:a9d4:24c8:0000
sys_image_guid: 14dd:a9d4:24c8:0003
vendor_id: 0x02c9
vendor_part_id: 4099
hw_ver: 0x0
board_id: AS_CX3_FDR
phys_port_cnt: 1
port: 1
state: PORT_ACTIVE (4)
max_mtu: 4096 (5)
active_mtu: 4096 (5)
sm_lid: 1
port_lid: 1
port_lmc: 0x00
link_layer: InfiniBand
cxgb4是以太网模块,iw_cxgb4是启用iwarp,只有iw_cxgb4模块加载成功iwarp功能才正常,判断iwarp是否启用可通过ibv_devices或ibv_devinfo 命令查看,这两条命令列出的都是ib设备。我这里有一张单端口的Mellonax FDR和双端口Chelsio网卡。
上面配置都完成后,才具备lustre安装基本环境。这里使用的 是lustre源码包安装,先编译生成rpm包后才能安装。
给lustre打补丁
这里chelsio提供了一个lustre_kernel.patch。官方文档中说需先打这个补丁才能正常工作。将lustre_kernel.path拷贝到lustre-2.5.41-ddn3-23bc966/lustre/kernel_patches/patches目录下,然后到lustre-2.5.41-ddn3-23bc966/lustre/kernel_patches/series目录下修改配置文件
[root@hpc-242-243 series]# vim 2.6-rhel6.6.series
lustre_version.patch
mpt-fusion-max-sge-rhel6.patch
raid5-mmp-unplug-dev-rhel6.patch
dev_read_only-2.6.32-rhel6.patch
blkdev_tunables-2.6-rhel6.patch
export-2.6.32-vanilla.patch
jbd2-jcberr-2.6-rhel6.patch
bh_lru_size_increase.patch
quota-replace-dqptr-sem.patch
quota-avoid-dqget-calls.patch
jbd2-log_wait_for_space-2.6-rhel6.patch
module-load-deadlock-rhel6.patch
dss-2.6.32-lustre-rhel6.6.patch
lustre_kernel.patch
在最后增加一行lustre_kernel.patch。
编译
[root@hpc-242-243 lustre-2.5.41-ddn3-23bc966]#./configure --with-linux=/usr/src/kernels/2.6.32-504.el6.x86_64 --with-o2ib=/usr/src/compat-rdma
[root@hpc-242-243 lustre-2.5.41-ddn3-23bc966]# make -j20
[root@hpc-242-243 lustre-2.5.41-ddn3-23bc966]# make rpms
完了之后在/root目录下会生成相关rpm安装包。
[root@hpc-242-243 x86_64]# ls
lustre-2.5.41-ddn3_2.6.32_504.3.3.el6_lustre.es131.x86_64_g23bc966.x86_64.rpm
lustre-debuginfo-2.5.41-ddn3_2.6.32_504.3.3.el6_lustre.es131.x86_64_g23bc966.x86_64.rpm
lustre-iokit-2.5.41-ddn3_2.6.32_504.3.3.el6_lustre.es131.x86_64_g23bc966.x86_64.rpm
lustre-modules-2.5.41-ddn3_2.6.32_504.3.3.el6_lustre.es131.x86_64_g23bc966.x86_64.rpm
lustre-osd-ldiskfs-2.5.41-ddn3_2.6.32_504.3.3.el6_lustre.es131.x86_64_g23bc966.x86_64.rpm
lustre-osd-ldiskfs-mount-2.5.41-ddn3_2.6.32_504.3.3.el6_lustre.es131.x86_64_g23bc966.x86_64.rpm
lustre-source-2.5.41-ddn3_2.6.32_504.3.3.el6_lustre.es131.x86_64_g23bc966.x86_64.rpm
lustre-tests-2.5.41-ddn3_2.6.32_504.3.3.el6_lustre.es131.x86_64_g23bc966.x86_64.rpm
[root@hpc-242-243 x86_64]# pwd
/root/rpmbuild/RPMS/x86_64
将这些rpm包在两个server节点上全部安装上。
修改lustre.conf文件,使用tcp还是iwarp网络,这里的写法很重要,这里是使用iwarp的方式。
[root@hpc-242-243 x86_64]# cat /etc/modprobe.d/lustre.conf
options lnet networks=o2ib0(eth3),tcp0(eth3)
如果是tcp,比较简单:
[root@hpc-242-243 x86_64]# cat /etc/modprobe.d/lustre.conf
options lnet networks=tcp(eth3)
将配置文件同步到另外一个IO节点上,加载lnet和lustre模块。
[root@hpc-242-243 x86_64]# modprobe lnet
[root@hpc-242-243 x86_64]# modprobe lustre
加载成功后,开始格式化磁盘,创建lustre文件系统
mkfs.lustre --mgs --mdt --fsname=lustre --mkfsoptions="-m 1 -J size=4096" --reformat --verbose /dev/mapper/360001ff0a003f000000000308a35000f ##创建mdt
格式化ost
mkfs.lustre --ost --index=0 --fsname=lustre --mgsnode=192.168.1.243@o2ib0 --mkfsoptions="-m 1 -i 131072" --reformat --writeconf --verbose /dev/mapper/360001ff0a003f000000000278a2c0006
mkfs.lustre --ost --index=1 --fsname=lustre --mgsnode=192.168.1.243@o2ib0 --mkfsoptions="-m 1 -i 131072" --reformat --writeconf --verbose /dev/mapper/360001ff0a003f000000000298a2e0008
mkfs.lustre --ost --index=2 --fsname=lustre --mgsnode=192.168.1.243@o2ib0 --mkfsoptions="-m 1 -i 131072" --reformat --writeconf --verbose /dev/mapper/360001ff0a003f000000000288a2d0007
mkfs.lustre --ost --index=3 --fsname=lustre --mgsnode=192.168.1.243@o2ib0 --mkfsoptions="-m 1 -i 131072" --reformat --writeconf --verbose /dev/mapper/360001ff0a003f0000000002a8a2f0009
mkfs.lustre --ost --index=4 --fsname=lustre --mgsnode=192.168.1.243@o2ib0 --mkfsoptions="-m 1 -i 131072" --reformat --writeconf --verbose /dev/mapper/360001ff0a003f0000000002b8a30000a
mkfs.lustre --ost --index=5 --fsname=lustre --mgsnode=192.168.1.243@o2ib0 --mkfsoptions="-m 1 -i 131072" --reformat --writeconf --verbose /dev/mapper/360001ff0a003f000000000318a360010
格式化完之后挂载
mount -t lustre /dev/mapper/360001ff0a003f000000000308a35000f /lustre/mdt
mount -t lustre /dev/mapper/360001ff0a003f000000000278a2c0006 /lustre/ost00
mount -t lustre /dev/mapper/360001ff0a003f000000000298a2e0008 /lustre/ost01
mount -t lustre /dev/mapper/360001ff0a003f000000000288a2d0007 /lustre/ost02
mount -t lustre /dev/mapper/360001ff0a003f000000000318a360010 /lustre/ost03
mount -t lustre /dev/mapper/360001ff0a003f0000000002a8a2f0009 /lustre/ost04
mount -t lustre /dev/mapper/360001ff0a003f0000000002b8a30000a /lustre/ost05
客户端的安装基本上和server相同,都需要安装ofed、网卡驱动,加载cxgb4和iw_cxgb4,只有这些都正确完成后才可以进行lustre安装。
首先编译lustre客户端安装包
按照前面的方式增加kernel补丁包之后进行编译生成rpm文件
[root@hpc-242-243 lustre-2.5.41-ddn3-23bc966]#./configure --with-linux=/usr/src/kernels/2.6.32-504.el6.x86_64 --with-o2ib=/usr/src/compat-rdma --disable-server
[root@hpc-242-243 lustre-2.5.41-ddn3-23bc966]# make -j20
[root@hpc-242-243 lustre-2.5.41-ddn3-23bc966]# make rpms
完了之后在/root目录下会生成相关rpm安装包
[root@hpc-242-242 x86_64]# ls
lustre-client-2.5.41-ddn3_2.6.32_504.el6.x86_64_g23bc966.x86_64.rpm
lustre-client-modules-2.5.41-ddn3_2.6.32_504.el6.x86_64_g23bc966.x86_64.rpm
lustre-client-source-2.5.41-ddn3_2.6.32_504.el6.x86_64_g23bc966.x86_64.rpm
lustre-client-tests-2.5.41-ddn3_2.6.32_504.el6.x86_64_g23bc966.x86_64.rpm
lustre-iokit-2.5.41-ddn3_2.6.32_504.el6.x86_64_g23bc966.x86_64.rpm
将这些安装包全部安装上,然后同样需要修改lustre.conf文件,配置同server上配置一样。
都安装完成后,加载相应模块,这里总结下需要加载的模块有4个
[root@hpc-242-242]#modprobe cxgb4 ##Chelsio网卡模块
[root@hpc-242-242]#modprobe iw_cxgb4 ##加载iwarp
[root@hpc-242-242]#modprobe lnet ##加载lustre网络
[root@hpc-242-242]#modprobe lustre ##加载lustre文件系统
这4个模块都正常加载成功,iwarp才能正常工作,他们之间是有一定的依赖,解决好这些依赖关系,才能顺利加载成功。
挂载
mount -t lustre -o localflock 192.168.1.243@o2ib:/lustre /lustre
挂载成功后即可开始性能测试了,这里使用的是iozone工具,我分别测试了不同数据块大小下的顺序读写性能,这里直接贴下测试脚本,具体过程就不详细说了。
[root@hpc-242-242 lustre]# cat run.sh
#!/bin/bash
for i in {4,8,16,64,1024}
do
/lustre/iozone/bin/iozone -i 0 -i 1 -r $i\k -s 20g -t 16 -+n -+m list
done
以上是加载iwarp的测试过程,tcp下比较简单,安装正常流程把chelsio网卡驱动安装后加载cxgb4模块后在安装lustre就行。这里不做详细说明了。下面贴下最终测试结果: