| 1 |
=================== |
| 2 |
Setting up NFS/RDMA |
| 3 |
=================== |
| 4 |
|
| 5 |
:Author: |
| 6 |
NetApp and Open Grid Computing (May 29, 2008) |
| 7 |
|
| 8 |
.. warning:: |
| 9 |
This document is probably obsolete. |
| 10 |
|
| 11 |
Overview |
| 12 |
======== |
| 13 |
|
| 14 |
This document describes how to install and setup the Linux NFS/RDMA client |
| 15 |
and server software. |
| 16 |
|
| 17 |
The NFS/RDMA client was first included in Linux 2.6.24. The NFS/RDMA server |
| 18 |
was first included in the following release, Linux 2.6.25. |
| 19 |
|
| 20 |
In our testing, we have obtained excellent performance results (full 10Gbit |
| 21 |
wire bandwidth at minimal client CPU) under many workloads. The code passes |
| 22 |
the full Connectathon test suite and operates over both Infiniband and iWARP |
| 23 |
RDMA adapters. |
| 24 |
|
| 25 |
Getting Help |
| 26 |
============ |
| 27 |
|
| 28 |
If you get stuck, you can ask questions on the |
| 29 |
[email protected] mailing list. |
| 30 |
|
| 31 |
Installation |
| 32 |
============ |
| 33 |
|
| 34 |
These instructions are a step by step guide to building a machine for |
| 35 |
use with NFS/RDMA. |
| 36 |
|
| 37 |
- Install an RDMA device |
| 38 |
|
| 39 |
Any device supported by the drivers in drivers/infiniband/hw is acceptable. |
| 40 |
|
| 41 |
Testing has been performed using several Mellanox-based IB cards, the |
| 42 |
Ammasso AMS1100 iWARP adapter, and the Chelsio cxgb3 iWARP adapter. |
| 43 |
|
| 44 |
- Install a Linux distribution and tools |
| 45 |
|
| 46 |
The first kernel release to contain both the NFS/RDMA client and server was |
| 47 |
Linux 2.6.25 Therefore, a distribution compatible with this and subsequent |
| 48 |
Linux kernel release should be installed. |
| 49 |
|
| 50 |
The procedures described in this document have been tested with |
| 51 |
distributions from Red Hat's Fedora Project (http://fedora.redhat.com/). |
| 52 |
|
| 53 |
- Install nfs-utils-1.1.2 or greater on the client |
| 54 |
|
| 55 |
An NFS/RDMA mount point can be obtained by using the mount.nfs command in |
| 56 |
nfs-utils-1.1.2 or greater (nfs-utils-1.1.1 was the first nfs-utils |
| 57 |
version with support for NFS/RDMA mounts, but for various reasons we |
| 58 |
recommend using nfs-utils-1.1.2 or greater). To see which version of |
| 59 |
mount.nfs you are using, type: |
| 60 |
|
| 61 |
.. code-block:: sh |
| 62 |
|
| 63 |
$ /sbin/mount.nfs -V |
| 64 |
|
| 65 |
If the version is less than 1.1.2 or the command does not exist, |
| 66 |
you should install the latest version of nfs-utils. |
| 67 |
|
| 68 |
Download the latest package from: https://www.kernel.org/pub/linux/utils/nfs |
| 69 |
|
| 70 |
Uncompress the package and follow the installation instructions. |
| 71 |
|
| 72 |
If you will not need the idmapper and gssd executables (you do not need |
| 73 |
these to create an NFS/RDMA enabled mount command), the installation |
| 74 |
process can be simplified by disabling these features when running |
| 75 |
configure: |
| 76 |
|
| 77 |
.. code-block:: sh |
| 78 |
|
| 79 |
$ ./configure --disable-gss --disable-nfsv4 |
| 80 |
|
| 81 |
To build nfs-utils you will need the tcp_wrappers package installed. For |
| 82 |
more information on this see the package's README and INSTALL files. |
| 83 |
|
| 84 |
After building the nfs-utils package, there will be a mount.nfs binary in |
| 85 |
the utils/mount directory. This binary can be used to initiate NFS v2, v3, |
| 86 |
or v4 mounts. To initiate a v4 mount, the binary must be called |
| 87 |
mount.nfs4. The standard technique is to create a symlink called |
| 88 |
mount.nfs4 to mount.nfs. |
| 89 |
|
| 90 |
This mount.nfs binary should be installed at /sbin/mount.nfs as follows: |
| 91 |
|
| 92 |
.. code-block:: sh |
| 93 |
|
| 94 |
$ sudo cp utils/mount/mount.nfs /sbin/mount.nfs |
| 95 |
|
| 96 |
In this location, mount.nfs will be invoked automatically for NFS mounts |
| 97 |
by the system mount command. |
| 98 |
|
| 99 |
.. note:: |
| 100 |
mount.nfs and therefore nfs-utils-1.1.2 or greater is only needed |
| 101 |
on the NFS client machine. You do not need this specific version of |
| 102 |
nfs-utils on the server. Furthermore, only the mount.nfs command from |
| 103 |
nfs-utils-1.1.2 is needed on the client. |
| 104 |
|
| 105 |
- Install a Linux kernel with NFS/RDMA |
| 106 |
|
| 107 |
The NFS/RDMA client and server are both included in the mainline Linux |
| 108 |
kernel version 2.6.25 and later. This and other versions of the Linux |
| 109 |
kernel can be found at: https://www.kernel.org/pub/linux/kernel/ |
| 110 |
|
| 111 |
Download the sources and place them in an appropriate location. |
| 112 |
|
| 113 |
- Configure the RDMA stack |
| 114 |
|
| 115 |
Make sure your kernel configuration has RDMA support enabled. Under |
| 116 |
Device Drivers -> InfiniBand support, update the kernel configuration |
| 117 |
to enable InfiniBand support [NOTE: the option name is misleading. Enabling |
| 118 |
InfiniBand support is required for all RDMA devices (IB, iWARP, etc.)]. |
| 119 |
|
| 120 |
Enable the appropriate IB HCA support (mlx4, mthca, ehca, ipath, etc.) or |
| 121 |
iWARP adapter support (amso, cxgb3, etc.). |
| 122 |
|
| 123 |
If you are using InfiniBand, be sure to enable IP-over-InfiniBand support. |
| 124 |
|
| 125 |
- Configure the NFS client and server |
| 126 |
|
| 127 |
Your kernel configuration must also have NFS file system support and/or |
| 128 |
NFS server support enabled. These and other NFS related configuration |
| 129 |
options can be found under File Systems -> Network File Systems. |
| 130 |
|
| 131 |
- Build, install, reboot |
| 132 |
|
| 133 |
The NFS/RDMA code will be enabled automatically if NFS and RDMA |
| 134 |
are turned on. The NFS/RDMA client and server are configured via the hidden |
| 135 |
SUNRPC_XPRT_RDMA config option that depends on SUNRPC and INFINIBAND. The |
| 136 |
value of SUNRPC_XPRT_RDMA will be: |
| 137 |
|
| 138 |
#. N if either SUNRPC or INFINIBAND are N, in this case the NFS/RDMA client |
| 139 |
and server will not be built |
| 140 |
|
| 141 |
#. M if both SUNRPC and INFINIBAND are on (M or Y) and at least one is M, |
| 142 |
in this case the NFS/RDMA client and server will be built as modules |
| 143 |
|
| 144 |
#. Y if both SUNRPC and INFINIBAND are Y, in this case the NFS/RDMA client |
| 145 |
and server will be built into the kernel |
| 146 |
|
| 147 |
Therefore, if you have followed the steps above and turned no NFS and RDMA, |
| 148 |
the NFS/RDMA client and server will be built. |
| 149 |
|
| 150 |
Build a new kernel, install it, boot it. |
| 151 |
|
| 152 |
Check RDMA and NFS Setup |
| 153 |
======================== |
| 154 |
|
| 155 |
Before configuring the NFS/RDMA software, it is a good idea to test |
| 156 |
your new kernel to ensure that the kernel is working correctly. |
| 157 |
In particular, it is a good idea to verify that the RDMA stack |
| 158 |
is functioning as expected and standard NFS over TCP/IP and/or UDP/IP |
| 159 |
is working properly. |
| 160 |
|
| 161 |
- Check RDMA Setup |
| 162 |
|
| 163 |
If you built the RDMA components as modules, load them at |
| 164 |
this time. For example, if you are using a Mellanox Tavor/Sinai/Arbel |
| 165 |
card: |
| 166 |
|
| 167 |
.. code-block:: sh |
| 168 |
|
| 169 |
$ modprobe ib_mthca |
| 170 |
$ modprobe ib_ipoib |
| 171 |
|
| 172 |
If you are using InfiniBand, make sure there is a Subnet Manager (SM) |
| 173 |
running on the network. If your IB switch has an embedded SM, you can |
| 174 |
use it. Otherwise, you will need to run an SM, such as OpenSM, on one |
| 175 |
of your end nodes. |
| 176 |
|
| 177 |
If an SM is running on your network, you should see the following: |
| 178 |
|
| 179 |
.. code-block:: sh |
| 180 |
|
| 181 |
$ cat /sys/class/infiniband/driverX/ports/1/state |
| 182 |
4: ACTIVE |
| 183 |
|
| 184 |
where driverX is mthca0, ipath5, ehca3, etc. |
| 185 |
|
| 186 |
To further test the InfiniBand software stack, use IPoIB (this |
| 187 |
assumes you have two IB hosts named host1 and host2): |
| 188 |
|
| 189 |
.. code-block:: sh |
| 190 |
|
| 191 |
host1$ ip link set dev ib0 up |
| 192 |
host1$ ip address add dev ib0 a.b.c.x |
| 193 |
host2$ ip link set dev ib0 up |
| 194 |
host2$ ip address add dev ib0 a.b.c.y |
| 195 |
host1$ ping a.b.c.y |
| 196 |
host2$ ping a.b.c.x |
| 197 |
|
| 198 |
For other device types, follow the appropriate procedures. |
| 199 |
|
| 200 |
- Check NFS Setup |
| 201 |
|
| 202 |
For the NFS components enabled above (client and/or server), |
| 203 |
test their functionality over standard Ethernet using TCP/IP or UDP/IP. |
| 204 |
|
| 205 |
NFS/RDMA Setup |
| 206 |
============== |
| 207 |
|
| 208 |
We recommend that you use two machines, one to act as the client and |
| 209 |
one to act as the server. |
| 210 |
|
| 211 |
One time configuration: |
| 212 |
----------------------- |
| 213 |
|
| 214 |
- On the server system, configure the /etc/exports file and start the NFS/RDMA server. |
| 215 |
|
| 216 |
Exports entries with the following formats have been tested:: |
| 217 |
|
| 218 |
/vol0 192.168.0.47(fsid=0,rw,async,insecure,no_root_squash) |
| 219 |
/vol0 192.168.0.0/255.255.255.0(fsid=0,rw,async,insecure,no_root_squash) |
| 220 |
|
| 221 |
The IP address(es) is(are) the client's IPoIB address for an InfiniBand |
| 222 |
HCA or the client's iWARP address(es) for an RNIC. |
| 223 |
|
| 224 |
.. note:: |
| 225 |
The "insecure" option must be used because the NFS/RDMA client does |
| 226 |
not use a reserved port. |
| 227 |
|
| 228 |
Each time a machine boots: |
| 229 |
-------------------------- |
| 230 |
|
| 231 |
- Load and configure the RDMA drivers |
| 232 |
|
| 233 |
For InfiniBand using a Mellanox adapter: |
| 234 |
|
| 235 |
.. code-block:: sh |
| 236 |
|
| 237 |
$ modprobe ib_mthca |
| 238 |
$ modprobe ib_ipoib |
| 239 |
$ ip li set dev ib0 up |
| 240 |
$ ip addr add dev ib0 a.b.c.d |
| 241 |
|
| 242 |
.. note:: |
| 243 |
Please use unique addresses for the client and server! |
| 244 |
|
| 245 |
- Start the NFS server |
| 246 |
|
| 247 |
If the NFS/RDMA server was built as a module (CONFIG_SUNRPC_XPRT_RDMA=m in |
| 248 |
kernel config), load the RDMA transport module: |
| 249 |
|
| 250 |
.. code-block:: sh |
| 251 |
|
| 252 |
$ modprobe svcrdma |
| 253 |
|
| 254 |
Regardless of how the server was built (module or built-in), start the |
| 255 |
server: |
| 256 |
|
| 257 |
.. code-block:: sh |
| 258 |
|
| 259 |
$ /etc/init.d/nfs start |
| 260 |
|
| 261 |
or |
| 262 |
|
| 263 |
.. code-block:: sh |
| 264 |
|
| 265 |
$ service nfs start |
| 266 |
|
| 267 |
Instruct the server to listen on the RDMA transport: |
| 268 |
|
| 269 |
.. code-block:: sh |
| 270 |
|
| 271 |
$ echo rdma 20049 > /proc/fs/nfsd/portlist |
| 272 |
|
| 273 |
- On the client system |
| 274 |
|
| 275 |
If the NFS/RDMA client was built as a module (CONFIG_SUNRPC_XPRT_RDMA=m in |
| 276 |
kernel config), load the RDMA client module: |
| 277 |
|
| 278 |
.. code-block:: sh |
| 279 |
|
| 280 |
$ modprobe xprtrdma.ko |
| 281 |
|
| 282 |
Regardless of how the client was built (module or built-in), use this |
| 283 |
command to mount the NFS/RDMA server: |
| 284 |
|
| 285 |
.. code-block:: sh |
| 286 |
|
| 287 |
$ mount -o rdma,port=20049 <IPoIB-server-name-or-address>:/<export> /mnt |
| 288 |
|
| 289 |
To verify that the mount is using RDMA, run "cat /proc/mounts" and check |
| 290 |
the "proto" field for the given mount. |
| 291 |
|
| 292 |
Congratulations! You're using NFS/RDMA! |