1 *************************************************************************
3 * Myrinet Express Lustre Networking Driver (MXLND) documentation *
5 *************************************************************************
9 MXLND provides support for Myricom's Myrinet Express (MX) communication
12 MXLND may be used with either MX-10G or MX-2G. See MX's README for
17 1. Configuring and compiling
21 1. Systems with different page sizes
23 3. MX endpoint collision
31 MXLND is supported on Linux 2.6. It may be possible to run it on 2.4,
32 but it has not been tested. MXLND requires Myricom's MX version 1.2.1
33 or higher. See MX's README for the supported list of processors.
35 1. Configuring and compiling
37 MXLND should be already integrated into the Lustre build process. To
38 build MXLND, you will need to set the path to your MX installation
39 in Lustre's ./configure:
43 replacing /opt with the actual path. Configure will check to ensure that
44 the MX version has the required functions. If not, it will fail to build.
45 To check if MXLND built, look for:
47 checking whether to enable Myrinet MX support... yes
49 in configure's output or the presence of Makefile in
50 $LUSTRE/lnet/klnds/mxlnd.
54 MXLND supports a number of load-time parameters using Linux's module
55 parameter system. On our test systems, we created the following file:
57 /etc/modprobe.d/kmxlnd
59 On some (older?) systems, you may need to modify /etc/modprobe.conf.
61 The available options are:
63 n_waitd # of completion daemons
64 max_peers maximum number of peers that may connect
65 cksum set non-zero to enable small message (< 4KB) checksums
66 ntx # of total tx message descriptors
67 credits # concurrent sends to a single peer
68 board index value of the Myrinet board (NIC)
70 polling Use 0 to block (wait). A value > 0 will poll that many times before blocking
71 hosts IP-to-hostname resolution file
73 Of these, only hosts is required. It must be the absolute path to the
74 MXLND hosts file. For example:
76 options kmxlnd hosts=/etc/hosts.mxlnd
78 The file format for the hosts file is as follows:
82 The values must be space and/or tab separated where:
84 IP is a valid IPv4 address
85 HOST is the name returned by `hostname` on that machine
86 BOARD is the index of the Myricom NIC (0 for the first card, etc.)
87 EP_ID is the MX endpoint ID
89 You may want to vary the remaining options to obtain the optimal performance
92 n_waitd sets the number of threads that process completed MX requests
93 (sends and receives). In our testing, the default of 1 performed best.
95 max_peers tells MXLND the upper limit of machines that it will need to
96 communicate with. This affects how many receives it will pre-post and each
97 receive will use one page of memory. Ideally, on clients, this value will
98 be equal to the total number of Lustre servers (MDS and OSS). On servers,
99 it needs to equal the total number of machines in the storage system.
101 cksum turns on small message checksums. It can be used to aid in trouble-
102 shooting. MX also provides an optional checksumming feature which can check
103 all messages (large and small). See the MX README for details.
105 ntx is the number of total sends in flight from this machine. In actuality,
106 MXLND reserves half of them for connect messages so make this value twice as large
107 as you want for the total number of sends in flight.
109 credits is the number of in-flight messages for a specific peer. This is part
110 of the flow-control system in Lustre. Increasing this value may improve performance
111 but it requires more memory since each message requires at least one page.
113 board is the index of the Myricom NIC. Hosts can have multiple Myricom NICs
114 and this identifies which one MXLND should use. This value must match the board
115 value in your MXLND hosts file for this host.
117 ep_id is the MX endpoint ID. Each process that uses MX is required to have at
118 least one MX endpoint to access the MX library and NIC. The ID is a simple index
119 starting at 0. This value must match the endpoint ID value in your MXLND hosts
122 polling determines whether this host will poll or block for MX request com-
123 pletions. A value of 0 blocks and any positive value will poll that many times
124 before blocking. Since polling increases CPU usage, we suggest you set this to
125 0 on the client and experiment with different values for servers.
127 =====================
128 II. MXLND Performance
129 =====================
131 On MX-2G systems, MXLND should easily saturate the link and use minimal CPU
132 (5-10% for read and write operations). On MX-10G systems, MXLND can saturate
133 the link and use moderate CPU resources (20-30% for read and write operations).
134 MX-10G relies on PCI-Express which is relatively new and performance varies
135 considerably by processor, motherboard and PCI-E chipset. Refer to Myricom's
136 website for the latest DMA read/write performance results by motherboard. The
137 DMA results will place an upper-bound on MXLND performance.
143 1. Systems with different page sizes
145 MXLND will set the maximum small message size equal to the kernel's page size.
146 This means that machines running MXLND that have different page sizes are not
147 able to communicate with each other. If you wish to run MXLND in this case,
148 send email to help@myri.com.
152 At this time, the MXLND cannot drive more than one interface at a time. Thus,
153 a single Lustre router cannot route between two MX-10G, between two MX-2G, or
154 between MX-10G and MX-2G fabrics.
156 3. MX endpoint collision
158 Each process that uses MX is required to have at least one MX endpoint to
159 access the MX library and NIC. Other processes may need to use MX and no two
160 processes can use the same endpoint ID. MPICH-MX dynamically chooses one at
161 MPI startup and should not interfere with MXLND. Sockets-MX, on the other hand,
162 is hard coded to use 0 for its ID. If it is possible that anyone will want to
163 run Sockets-MX on this system, use a non-0 value for MXLND's endpoint ID.
170 MXLND is copyright (C) 2006 of Myricom, Inc.
172 MXLND is part of Lustre, http://www.lustre.org.
174 MXLND is free software; you can redistribute it and/or modify it under the
175 terms of version 2 of the GNU General Public License as published by the Free
178 MXLND is distributed in the hope that it will be useful, but WITHOUT ANY
179 WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A
180 PARTICULAR PURPOSE. See the GNU General Public License for more details.
182 You should have received a copy of the GNU General Public License along with
183 Lustre; if not, write to the Free Software Foundation, Inc., 675 Mass Ave,
184 Cambridge, MA 02139, USA.
190 If you have questions about MXLND, please contact help@myri.com.