| 1 |
README/Release Notes |
| 2 |
OFED 4.8 DAPL Release 2.1.10 |
| 3 |
Dec 2016 |
| 4 |
|
| 5 |
User space libraries/utilities for Direct Access Transport (DAT) v2.0. DAT is |
| 6 |
a transport-independent, platform-independent Application Programming |
| 7 |
Interface that supports RDMA (remote direct memory access) devices. |
| 8 |
Note: v1.2 is no longer supported and will not be included with OFED releases |
| 9 |
|
| 10 |
MIC support is provided with the new MCM provider and MPXYD service, since dapl-2.1.0. |
| 11 |
MCM requires the Intel(R) MPSS 3.x (YOCTO) release for Linux to be installed on your system. |
| 12 |
MPSS 3.x for Linux can be downloaded from: http://software.intel.com/mic-developer |
| 13 |
|
| 14 |
For latest documentation and packages: //www.openfabrics.org/downloads/dapl/ |
| 15 |
|
| 16 |
================= |
| 17 |
1.0 Release Notes |
| 18 |
================= |
| 19 |
dapl-2.1.10 changes include bug fixes, CCL Proxy post_send optimizations, dtest/dtest_suite improvements |
| 20 |
|
| 21 |
dapl-2.1.9-2 changes include bug fixes, MFO fixes, GID query attributes |
| 22 |
dapl-2.1.9-1 changes include dtestcm improvement, bug fixes for MIC, Fix SCM interop issue with MTU |
| 23 |
|
| 24 |
dapl-2.1.8 changes include dtest improvement, adding Intel OPA support, and MTU negotiation |
| 25 |
- Add dtest -D option for data check/validation via pingpong test |
| 26 |
- Add dtest -q option for open/query only option |
| 27 |
- Add support for new hfi (Intel OPA) driver, all providers |
| 28 |
CCL Proxy can support hfi in MFO mode via mcm provider |
| 29 |
- Active MTU is now set as default QP MTU, CM mechanism added to negotiate |
| 30 |
and fallback to smaller size of endpoints. Is backward compatible with |
| 31 |
older providers running defaults of 2K and can be adjusted with |
| 32 |
pre-existing DAPL_IB_MTU environment setting. |
| 33 |
- Bug fixes, including CCL Data corruption (scif_writeto ordering requirements). |
| 34 |
- Provide new test/dtest/dtest_suite.sh script for automated host to mic test/validation. |
| 35 |
|
| 36 |
dapl-2.1.7 changes include dtest improvement and CCL Proxy P2P inline support: |
| 37 |
|
| 38 |
- Add dtest -W option for rdma write pinpong test, |
| 39 |
new options with -W include -a (all data sizes) -i (incremental size) |
| 40 |
- CCL Proxy small message latency improvment with Proxy2Proxy inline support |
| 41 |
for message sizes < 96 bytes, reduces MPI pingpong single byte latency |
| 42 |
for MFO devices by 27%. |
| 43 |
|
| 44 |
dapl-2.1.6 changes include MIC support for full offload mode |
| 45 |
|
| 46 |
- Add support for Truescale qib devices with no CCL Direct verbs support on MIC. |
| 47 |
- Enhancement for inside the box transfers without IB adapter via ibscif. |
| 48 |
- Add DAPL_NETWORK_NODES, DAPL_NETWORK_PPN environment variables. |
| 49 |
|
| 50 |
dapl-2.1.5 changes include improvements for large scale UD communication management: |
| 51 |
|
| 52 |
- AH caching, reduced memory footprint (grows as needed) |
| 53 |
- Port space increased to 24 bits |
| 54 |
- Hash table for port space, CM object management |
| 55 |
- Optimized CM wire protocol for fast index lookup |
| 56 |
|
| 57 |
Tested on 1200n 28ppn cluster, AlltoAll Intel MPI, UD mode. |
| 58 |
Both static and dynamic modes, over 500m UD QP connections. |
| 59 |
|
| 60 |
========== |
| 61 |
2.0 BUILD: |
| 62 |
========== |
| 63 |
|
| 64 |
# NON_DEBUG build/install example for x86_64, OFED targets |
| 65 |
./configure --prefix /usr --sysconf=/etc --libdir /usr/lib64 LDFLAGS=-L/usr/lib64 CPPFLAGS="-I/usr/include" |
| 66 |
make install |
| 67 |
|
| 68 |
# DEBUG build/install example for x86_64, using OFED targets |
| 69 |
./configure --enable-debug --prefix /usr --sysconf=/etc --libdir /usr/lib64 LDFLAGS=-L/usr/lib64 CPPFLAGS="-I/usr/include" |
| 70 |
make install |
| 71 |
|
| 72 |
# COUNTERS build/install example for x86_64, using OFED targets |
| 73 |
./configure --prefix /usr --sysconf=/etc --libdir /usr/lib64 LDFLAGS=-L/usr/lib64 CPPFLAGS="-I/usr/include -DDAPL_COUNTERS" |
| 74 |
make install |
| 75 |
|
| 76 |
========================================================= |
| 77 |
3.0 Provider descriptions and CM results (cma, scm, ucm): |
| 78 |
========================================================= |
| 79 |
|
| 80 |
1. CMA - uses OFA rdma_cm to setup QP's. IPoIB, ARP, and SA queries required. |
| 81 |
|
| 82 |
Provider name: ofa-v2-ib0 |
| 83 |
PROs: OFA rdma_cm has the most testing across many applications. |
| 84 |
Supports both iWARP and IB. |
| 85 |
|
| 86 |
CONs: Serialization of conn processing with kernel based CM service |
| 87 |
Requires IPoIB ARP for name resolution, storms |
| 88 |
Requires SA for path record queries for IB fabrics. |
| 89 |
Conn Request private data limited to 52 bytes. |
| 90 |
|
| 91 |
Settings for larger clusters (512+ cores): |
| 92 |
|
| 93 |
setenv DAPL_CM_ROUTE_TIMEOUT_MS 20000 |
| 94 |
setenv DAPL_CM_ARP_TIMEOUT_MS 10000 |
| 95 |
|
| 96 |
2. SCM - uses sockets to exchange QP information. IPoIB, ARP, and SA queries NOT required. |
| 97 |
|
| 98 |
Provider name (connectx): ofa-v2-mlx4_0-1 |
| 99 |
PROs: Each rank has own instance of socket cm. More private data with requests. |
| 100 |
Doesn't require path-record lookup. |
| 101 |
|
| 102 |
CONs: Socket resources grow with scale-out, serialization of |
| 103 |
connections with kernel based tcp sockets, |
| 104 |
Competes for MPI socket resources/port space and other TCP applications. |
| 105 |
Sockets remain in TIMEWAIT state for minutes after closure. |
| 106 |
Requires ARP for name resolution. |
| 107 |
Doesn't support iWARP devices. |
| 108 |
|
| 109 |
Settings for larger clusters (512+ cores): |
| 110 |
|
| 111 |
setenv DAPL_ACK_RETRY 7 /* IB RC Ack retry count */ |
| 112 |
setenv DAPL_ACK_TIMER 20 /* IB RC Ack retry timer */ |
| 113 |
|
| 114 |
3. UCM - use's IB UD QP to exchange QP info. Sockets, ARP, IPoIB, and SA queries NOT required. |
| 115 |
|
| 116 |
Provider name (connectx): ofa-v2-mlx4_0-1u |
| 117 |
PROs: Each rank has own instance of CM in user process |
| 118 |
Resources fixed per rank regardless of scale-out size |
| 119 |
No serialization of user or kernel resources establishing connections, |
| 120 |
Simple 3-way msg handsake, CM messages fit in inline data for lowest message latency, |
| 121 |
Supports alternate paths |
| 122 |
No address resolution required. |
| 123 |
No path resolution required. |
| 124 |
|
| 125 |
CONs: New provider with limited testing, a little tougher to debug. |
| 126 |
Doesn't support iWARP |
| 127 |
|
| 128 |
Settings for larger clusters (512+ cores): |
| 129 |
|
| 130 |
setenv DAPL_UCM_REP_TIME 10000 /* REQUEST timer, waiting for REPLY in millisecs */ |
| 131 |
setenv DAPL_UCM_RTU_TIME 10000 /* REPLY timer, waiting for RTU in millisecs */ |
| 132 |
setenv DAPL_UCM_CQ_SIZE 2000 /* CM completion queue */ |
| 133 |
setenv DAPL_UCM_QP_SIZE 2000 /* CM message queue */ |
| 134 |
setenv DAPL_UCM_RETRY 7 /* REQUEST and REPLY retries */ |
| 135 |
setenv DAPL_ACK_RETRY 7 /* IB RC Ack retry count */ |
| 136 |
setenv DAPL_ACK_TIMER 20 /* IB RC Ack retry timer */ |
| 137 |
|
| 138 |
CM Performance: CPS profile for cma, scm, and ucm v2 uDAPL providers: |
| 139 |
----------------------------------------------------------------------- |
| 140 |
Intel(R) Xeon(R) CPU E5-2690 v2 @ 3.00GHz (IVT) |
| 141 |
Mellanox MLX4 IB FDR, no switch. |
| 142 |
|
| 143 |
dtestcm (server/client): |
| 144 |
|
| 145 |
cma: Connections: 313.10 usec, CPS 3193.83 Total 0.31 secs, poll_cnt=6300, Num=1000 |
| 146 |
scm: Connections: 167.65 usec, CPS 5964.92 Total 0.17 secs, poll_cnt=2394, Num=1000 |
| 147 |
ucm: Connections: 71.85 usec, CPS 13918.06 Total 0.07 secs, poll_cnt=2360, Num=1000 |
| 148 |
|
| 149 |
dapl_cm_bw: MPI uDAPL/CM profiling application (all-to-all connections, all ranks) |
| 150 |
|
| 151 |
CMA |
| 152 |
2 Connect times (10): Total 0.0049 per 0.0005 CPS=2051.38 |
| 153 |
4 Connect times (40): Total 0.0151 per 0.0004 CPS=2650.16 |
| 154 |
8 Connect times (240): Total 0.0548 per 0.0002 CPS=4380.59 |
| 155 |
16 Connect times (1120): Total 4.0356 per 0.0036 CPS=277.53 |
| 156 |
32 Connect times (4800): Total 4.4704 per 0.0009 CPS=1073.72 |
| 157 |
|
| 158 |
SCM |
| 159 |
2 Connect times (10): Total 0.0029 per 0.0003 CPS=3441.31 |
| 160 |
4 Connect times (40): Total 0.0060 per 0.0002 CPS=6635.97 |
| 161 |
8 Connect times (240): Total 0.0194 per 0.0001 CPS=12383.47 |
| 162 |
16 Connect times (1120): Total 0.0649 per 0.0001 CPS=17246.93 |
| 163 |
32 Connect times (4800): Total 1.0193 per 0.0002 CPS=4708.95 |
| 164 |
|
| 165 |
UCM |
| 166 |
2 Connect times (10): Total 0.0014 per 0.0001 CPS=6993.91 |
| 167 |
4 Connect times (40): Total 0.0045 per 0.0001 CPS=8837.87 |
| 168 |
8 Connect times (240): Total 0.0155 per 0.0001 CPS=15477.13 |
| 169 |
16 Connect times (1120): Total 0.0630 per 0.0001 CPS=17765.12 |
| 170 |
32 Connect times (4800): Total 0.2632 per 0.0001 CPS=18236.54 |
| 171 |
|
| 172 |
=================================================================================================== |
| 173 |
4.0 BKM for installing new DAPL library on your cluster without any impact on existing OFED install: |
| 174 |
==================================================================================================== |
| 175 |
|
| 176 |
Note: example for user /home/user1, (assumes /home/user1 is exported) and MLX4 adapter, port 1 |
| 177 |
|
| 178 |
Download latest 2.1.x package: http://www.openfabrics.org/downloads/dapl/dapl-2.1.x.tar.gz |
| 179 |
|
| 180 |
untar in /home/user1 |
| 181 |
cd /home/user1/dapl-2.1.x |
| 182 |
./configure LDFLAGS=-L/usr/lib64 CPPFLAGS="-I/usr/include" |
| 183 |
make |
| 184 |
|
| 185 |
Create /home/user1/dat.conf with following 3 lines. (entries with path to new libraries): |
| 186 |
|
| 187 |
ofa-v2-mlx4_0-1u u2.0 nonthreadsafe default /home/user1/dapl-2.1.x/dapl/udapl/.libs/libdaploucm.so.2 dapl.2.0 "mlx4_0 1" "" |
| 188 |
ofa-v2-mlx4_0-1m u2.0 nonthreadsafe default /home/user1/dapl-2.1.x/dapl/udapl/.libs/libdaplomcm.so.2 dapl.2.0 "mlx4_0 1" "" |
| 189 |
ofa-v2-mlx4_0-1 u2.0 nonthreadsafe default /home/user1/dapl-2.1.x/dapl/udapl/.libs/libdaploscm.so.2 dapl.2.0 "mlx4_0 1" "" |
| 190 |
ofa-v2-ib0 u2.0 nonthreadsafe default /home/user1/dapl-2.1.x/dapl/udapl/.libs/libdaplcma.so.1 dapl.2.0 "ib0 0" "" |
| 191 |
|
| 192 |
Run uDAPL application or Intel MPI that uses uDAPL, with (assuming mlx4_0 adapters) following: |
| 193 |
|
| 194 |
setenv DAT_OVERRIDE=/home/user1/dat.conf |
| 195 |
setenv LD_LIBRARY_PATH=/home/user1/dapl-2.1.5/dapl/udapl/.libs:$LD_LIBRARY_PATH |
| 196 |
|
| 197 |
If running Intel MPI and uDAPL IB UD cm, set the following (recommended): |
| 198 |
|
| 199 |
setenv I_MPI_DAPL_PROVIDER=ofa-v2-mlx4_0-1u |
| 200 |
|
| 201 |
If running Intel MPI and uDAPL IB mcm with MIC, set the following: |
| 202 |
|
| 203 |
setenv I_MPI_DAPL_PROVIDER=ofa-v2-mlx4_0-1m |
| 204 |
|
| 205 |
If running Intel MPI and uDAPL socket cm, set the following: |
| 206 |
|
| 207 |
setenv I_MPI_DAPL_PROVIDER=ofa-v2-mlx4_0-1 |
| 208 |
|
| 209 |
|
| 210 |
If running Intel MPI and uDAPL rdma_cm, set the following: |
| 211 |
|
| 212 |
setenv I_MPI_DAPL_PROVIDER=ofa-v2-ib0 |
| 213 |
|
| 214 |
|
| 215 |
============================================================ |
| 216 |
5.0 MCM Provider, MPXYD Daemon (CCL-proxy) Build and Install |
| 217 |
============================================================ |
| 218 |
|
| 219 |
MCM is a new uDAPL provider that is an extension to standard DAT 2.0 libraries. The purpose of this service |
| 220 |
is to proxy RDMA writes from the MIC to the HOST to improve large IO performance. The provider will support |
| 221 |
MIC to MIC, HOST to HOST, and MIC to HOST environments. The mcm client will NOT use MPXYD when running on the host. |
| 222 |
It requires a new MPXYD daemon service when clients are running on a MIC KNC adapter. This package installs all the |
| 223 |
host side libraries and daemon service. The MIC libraries must be built and moved over to MIC adapter. This verion |
| 224 |
is currently included with MPSS and all libraries and services will be installed by default. |
| 225 |
|
| 226 |
Current release package: dapl-2.1.9.tar.gz |
| 227 |
|
| 228 |
* Sample host build from source package (ofed must be installed) |
| 229 |
|
| 230 |
./configure --enable-mcm --prefix=/usr --libdir=/usr/lib64 --sysconfdir=/etc |
| 231 |
make |
| 232 |
make install |
| 233 |
|
| 234 |
* Sample host rpmbuild/update from release tarball, /root: |
| 235 |
|
| 236 |
rpmbuild -ta dapl-2.1.9.tar.gz |
| 237 |
rpm -U /root/rpmbuild/RPMS/x86_64/dapl* |
| 238 |
|
| 239 |
* Sample MIC build from source package for MPSS 3.x KNC (MPSS must be installed) |
| 240 |
* Assume /opt is nfs mounted across cluster |
| 241 |
|
| 242 |
source /opt/mpss/3.x/environment-setup-k1om-mpss-linux |
| 243 |
./configure --enable-mcm --prefix /opt/dapl/mic --host=x86_64-k1om-linux |
| 244 |
make |
| 245 |
make install |
| 246 |
|
| 247 |
copy /opt/dapl/mic/* files out to all MIC cards |
| 248 |
|
| 249 |
* Cluster deployment |
| 250 |
|
| 251 |
(1) Build once on the head or on one of the nodes (with MPSS) as described in the above steps. |
| 252 |
|
| 253 |
(2) HOST: Install dapl libraries and mpxyd service, "rpm -U" all dapl RPM files on host nodes: |
| 254 |
|
| 255 |
(3) MIC: Setup dapl overlay for new package (/opt/intel/dapl): |
| 256 |
|
| 257 |
Create /etc/mpss/conf.d/dapl.conf with following entry: |
| 258 |
|
| 259 |
Overlay Filelist /opt/dapl /opt/dapl/dapl.filelist on |
| 260 |
|
| 261 |
Create /opt/dapl/dapl.filelist with following entries: |
| 262 |
|
| 263 |
file /etc/dat.conf mic/etc/dat.conf 755 0 0 |
| 264 |
file /usr/bin/dtest mic/bin/dtest 755 0 0 |
| 265 |
file /usr/bin/dtestx mic/bin/dtestx 755 0 0 |
| 266 |
file /usr/bin/dtestcm mic/bin/dtestcm 755 0 0 |
| 267 |
file /usr/bin/dapltest mic/bin/dapltest 755 0 0 |
| 268 |
file /usr/lib64/libdat.so.2.0.0 mic/lib/libdat.so.2.0.0 755 0 0 |
| 269 |
file /usr/lib64/libdaplofa.so.2.0.0 mic/lib/libdaplofa.so.2.0.0 755 0 0 |
| 270 |
file /usr/lib64/libdaplomcm.so.2.0.0 mic/lib/libdaplomcm.so.2.0.0 755 0 0 |
| 271 |
file /usr/lib64/libdaploscm.so.2.0.0 mic/lib/libdaploscm.so.2.0.0 755 0 0 |
| 272 |
file /usr/lib64/libdaploucm.so.2.0.0 mic/lib/libdaploucm.so.2.0.0 755 0 0 |
| 273 |
|
| 274 |
slink /usr/lib64/libdat.so libdat.so.2.0.0 777 0 0 |
| 275 |
slink /usr/lib64/libdat.so.2 libdat.so.2.0.0 777 0 0 |
| 276 |
slink /usr/lib64/libdaplofa.so libdaplofa.so.2.0.0 777 0 0 |
| 277 |
slink /usr/lib64/libdaplofa.so.2 libdaplofa.so.2.0.0 777 0 0 |
| 278 |
slink /usr/lib64/libdaplomcm.so libdaplomcm.so.2.0.0 777 0 0 |
| 279 |
slink /usr/lib64/libdaplomcm.so.2 libdaplomcm.so.2.0.0 777 0 0 |
| 280 |
slink /usr/lib64/libdaploscm.so libdaploscm.so.2.0.0 777 0 0 |
| 281 |
slink /usr/lib64/libdaploscm.so.2 libdaploscm.so.2.0.0 777 0 0 |
| 282 |
slink /usr/lib64/libdaploucm.so libdaploucm.so.2.0.0 777 0 0 |
| 283 |
slink /usr/lib64/libdaploucm.so.2 libdaploucm.so.2.0.0 777 0 0 |
| 284 |
|
| 285 |
Reboot or restart MPSS and ofed-mic services |
| 286 |
|
| 287 |
Check for dapl overlay |
| 288 |
micctrl --config |
| 289 |
|
| 290 |
* Setup for non-root CCL Proxy testing, MPXYD running as process with different service port from your /home directory: |
| 291 |
|
| 292 |
Using build instructions above, change prefix as follow and "make install": |
| 293 |
|
| 294 |
Build MIC: |
| 295 |
--prefix=/home/username/ccl-proxy-mic |
| 296 |
|
| 297 |
Build host: |
| 298 |
--prefix=/home/username/ccl-proxy-host |
| 299 |
|
| 300 |
edit /home/username/ccl-proxy-host/etc/mpxyd.conf and change the following entries: |
| 301 |
|
| 302 |
log_file /var/log/mpxyd.log to log_file /tmp/username/mpxyd.log |
| 303 |
lock_file /var/log/mpxyd.pid to lock_file /tmp/username/mpxyd.log |
| 304 |
scif_port_id 68 to scif_port_id 1068 |
| 305 |
|
| 306 |
start the mpxyd process on each node |
| 307 |
|
| 308 |
ssh node1-hostname /home/username/ccl-proxy-host/sbin/mpxyd -P -O /home/username/ccl-proxy-host/etc/mpxyd.conf& |
| 309 |
|
| 310 |
Note: override default port id using following environment variable: |
| 311 |
|
| 312 |
export DAPL_MCM_PORT_ID=1068 |
| 313 |
|
| 314 |
* Notes |
| 315 |
|
| 316 |
(1) Modify "/etc/mpxyd.conf" to change the settings for the proxy. Especially, try different values |
| 317 |
of "buffer_segment_size" for performance tuning. Use a smaller value for "buffer_pool_mb" |
| 318 |
to reduce the memory foorprint of mpxyd. Use a larger value for "scif_listen_qlen" to run |
| 319 |
more MPI ranks per card. Also modify mcm_affinity_base to the desired CPU_id to insure |
| 320 |
socket to adapter affinity. Best performance when HCA, MIC, and CPU are on same socket. |
| 321 |
Default settings are on CPU socket 0. |
| 322 |
|
| 323 |
(2) By default, only writes originated from MIC is proxied. However, it is also possible to proxy |
| 324 |
host-originated writes (e.g. for debugging purpose). To do this, set the environment variable |
| 325 |
"DAPL_MCM_ALWAYS_PROXY=1". This variable applies to the provider, not the proxy. |
| 326 |
|
| 327 |
(3) Use the MCM provider with Intel MPI 5.1 or greater for best out of box experiences with MIC. |
| 328 |
|
| 329 |
Recommended settings: |
| 330 |
|
| 331 |
export I_MPI_MIC=1 |
| 332 |
export I_MPI_DEBUG=2 |
| 333 |
export I_MPI_FALLBACK=0 |
| 334 |
|
| 335 |
(4) Server Platform Notes: |
| 336 |
|
| 337 |
For optimal CCL Proxy performance, installed on Intel® Xeon® Processor E5-26xx v2 and v3 platforms, |
| 338 |
please update to the latest platform BIOS and set the following Processor, Power, and Memory settings. |
| 339 |
|
| 340 |
https://downloadcenter.intel.com/default.aspx |
| 341 |
|
| 342 |
[BIOS::Advanced::Processor Configuration] |
| 343 |
Intel(R) QPI Frequency Select=Auto Max |
| 344 |
Intel(R) Turbo Boost Technology=Enabled |
| 345 |
Enhanced Intel SpeedStep(R) Tech=Enabled |
| 346 |
Processor C3=Disabled |
| 347 |
Processor C6=Enabled |
| 348 |
Intel(R) Virtualization Technology=Disabled |
| 349 |
Intel(R) VT for Directed I/O=Disabled |
| 350 |
Direct Cache Access (DCA)=Enabled |
| 351 |
Extended ATR=0x01 |
| 352 |
|
| 353 |
[BIOS::Advanced::Power & Performance] |
| 354 |
CPU Power and Performance Policy=Performance |
| 355 |
|
| 356 |
[BIOS::Advanced::Memory Configuration::Memory RAS and Performance Configuration] |
| 357 |
Select Memory RAS Configuration=Maximum Performance |
| 358 |
NUMA Optimized=Enabled |
| 359 |
|
| 360 |
==================================================================== |
| 361 |
6.0 Mellanox Fabric Collective Accelerator (FCA) build for Intel MPI |
| 362 |
==================================================================== |
| 363 |
|
| 364 |
1) Download latest package: http://www.openfabrics.org/downloads/dapl/dapl-2.x.x.tar.gz |
| 365 |
|
| 366 |
2) Build/Install with collectives configured for fca type: |
| 367 |
|
| 368 |
./configure --enable-coll-type=fca --disable-mcm --prefix=/usr |
| 369 |
LDFLAGS=-L/opt/mellanox/fca/lib |
| 370 |
CPPFLAGS=-I/opt/mellanox/fca/include |
| 371 |
make install |
| 372 |
|
| 373 |
If an uDAPL provider with collective support is installed as an additional package, |
| 374 |
while another uDAPL provider also exists in a system, LD_LIBRARY_PATH environment |
| 375 |
variable should be updated to include the path to the newly installed DAPL package. |
| 376 |
|
| 377 |
3) Set following MPI variables: |
| 378 |
|
| 379 |
export FCA_MGR_HOME=/opt/mellanox/fca |
| 380 |
export FCA_HOME=/opt/mellanox/fca |
| 381 |
export I_MPI_DAPL_COLLECTIVES=all |
| 382 |
export I_MPI_FABRICS=dapl |
| 383 |
export I_MPI_FALLBACK_DEVICE=off |
| 384 |
|
| 385 |
============================= |
| 386 |
7.0 Environment Variables |
| 387 |
============================= |
| 388 |
|
| 389 |
- IB UD options using UCM provider, large scale settings (Xeon) |
| 390 |
|
| 391 |
export DAPL_NETWORK_NODES= /* set to active nodes on network for CM */ |
| 392 |
export DAPL_NETWORK_PPN= /* set to active processes per node for CM */ |
| 393 |
|
| 394 |
/* The following will be adjusted by provider based on NODES, PPN */ |
| 395 |
export DAPL_UCM_REP_TIME=8000 /* REQUEST timer, waiting on REPLY, msecs, default = 800 */ |
| 396 |
export DAPL_UCM_RTU_TIME=8000 /* REPLY timer, waiting for RTU in msecs, default=400 */ |
| 397 |
export DAPL_UCM_RETRY=7 /* REQUEST & REPLY retries, default = 7 */ |
| 398 |
export DAPL_UCM_QP_SIZE=4000 /* CM req/reply work queue size, default = 500 entries */ |
| 399 |
export DAPL_UCM_CQ_SIZE=4000 /* CM req/reply completion queue size, default = 500 entries */ |
| 400 |
export DAPL_UCM_TX_BURST=100 /* CM signal rate on send messages */ |
| 401 |
export DAPL_UCM_ENTRY_BITS=11 /* default = 11-bit, 2KB entries, allocation blocks */; |
| 402 |
export DAPL_UCM_ARRAY_BITS=18 /* default = 18 bit, 256KB total */ |
| 403 |
|
| 404 |
- IB RC options using SCM provider |
| 405 |
|
| 406 |
export DAPL_SCM_NETDEV=ib0 /* default is first non-loopback netdev */ |
| 407 |
|
| 408 |
- Other IB settings for all providers: |
| 409 |
|
| 410 |
export DAPL_MAX_INLINE=64 /* IB RC inline optimization, best small msg latency, def=64 */ |
| 411 |
export DAPL_ACK_RETRY=7 /* IB RC Ack retry count, default 7 */ |
| 412 |
export DAPL_ACK_TIMER=20 /* IB RC Ack retry timer, 5 bits, 4.096us*2^ack_timer. 16== 268ms, 20==4.2s */ |
| 413 |
export DAPL_IB_MTU=2048 /* IB MTU size, default = 2048 */ |
| 414 |
export DAPL_RNR_TIMER=12 /* 5 bits, 12 =.64ms, 28 =163ms, 31 =491ms */ |
| 415 |
export DAPL_RNR_RETRY=7 /* 3 bits, 7 == infinite */ |
| 416 |
export DAPL_IB_PKEY= 0 /* override IB partition key, default is pkey index 0 */ |
| 417 |
export DAPL_IB_SL=0 /* override IB Sevice level, default = 0 */ |
| 418 |
|
| 419 |
- Other options: |
| 420 |
export DAPL_WR_MAX=500 /* used to reduce max qp depth on all IB providers, default = dev attributes */ |
| 421 |
|
| 422 |
Debug logging and Counter settings ( --enable-counters) |
| 423 |
|
| 424 |
export DAPL_DBG_SYS_MEM=10 /* threshold for low sys memory warning, def = 10 percent */ |
| 425 |
export DAPL_DBG_TYPE=0x0000003 /* set log, monitor, and error checking, default = warnings and errors */ |
| 426 |
|
| 427 |
DAPL_DBG_TYPE bit settings as follow: |
| 428 |
|
| 429 |
DAPL_DBG_TYPE_ERR = 0x0001, |
| 430 |
DAPL_DBG_TYPE_WARN = 0x0002, |
| 431 |
DAPL_DBG_TYPE_EVD = 0x0004, |
| 432 |
DAPL_DBG_TYPE_CM = 0x0008, |
| 433 |
DAPL_DBG_TYPE_EP = 0x0010, |
| 434 |
DAPL_DBG_TYPE_UTIL = 0x0020, |
| 435 |
DAPL_DBG_TYPE_CALLBACK = 0x0040, |
| 436 |
DAPL_DBG_TYPE_DTO_COMP_ERR = 0x0080, |
| 437 |
DAPL_DBG_TYPE_API = 0x0100, |
| 438 |
DAPL_DBG_TYPE_RTN = 0x0200, |
| 439 |
DAPL_DBG_TYPE_EXCEPTION = 0x0400, |
| 440 |
DAPL_DBG_TYPE_SRQ = 0x0800, |
| 441 |
DAPL_DBG_TYPE_CNTR = 0x1000, |
| 442 |
DAPL_DBG_TYPE_CM_LIST = 0x2000, |
| 443 |
DAPL_DBG_TYPE_THREAD = 0x4000, |
| 444 |
DAPL_DBG_TYPE_CM_EST = 0x8000, |
| 445 |
DAPL_DBG_TYPE_CM_WARN = 0x10000, |
| 446 |
DAPL_DBG_TYPE_EXTENSION = 0x20000, |
| 447 |
DAPL_DBG_TYPE_CM_STATS = 0x40000, |
| 448 |
DAPL_DBG_TYPE_CM_ERRS = 0x80000, /* print any cm errors on device close */ |
| 449 |
DAPL_DBG_TYPE_LINK_ERRS = 0x100000, /* print any link errors on device close */ |
| 450 |
DAPL_DBG_TYPE_LINK_WARN = 0x200000, /* print any link warning on device close */ |
| 451 |
DAPL_DBG_TYPE_DIAG_ERRS = 0x400000, /* print any diag_counter errors on dev close */ |
| 452 |
DAPL_DBG_TYPE_SYS_WARN = 0x800000, /* print low mem warning during alloc, reg_mem */ |
| 453 |
DAPL_DBG_TYPE_VER = 0x1000000, /* print dapl ver and build date during dev open */ |
| 454 |
|
| 455 |
============================= |
| 456 |
9.0 SAMPLE uDAPL APPLICATION: |
| 457 |
============================= |
| 458 |
|
| 459 |
There are 2 sample programs, with manpages, provided with this package. |
| 460 |
|
| 461 |
(dapl/test/dtest/) |
| 462 |
|
| 463 |
NAME |
| 464 |
dtest - simple uDAPL send/receive and RDMA test |
| 465 |
|
| 466 |
SYNOPSIS |
| 467 |
dtest [-P provider] [-b buf size] [-B burst count][-v] [-c] [-p] [-d] [-s] |
| 468 |
|
| 469 |
dtest [-P provider] [-b buf size] [-B burst count][-v] [-c] [-p] [-d] [-h HOSTNAME] |
| 470 |
|
| 471 |
DESCRIPTION |
| 472 |
dtest is a simple test used to exercise and verify the uDAPL interfaces. At least two instantia- |
| 473 |
tions of the test must be run. One acts as the server and the other the client. The server side of |
| 474 |
the test, once invoked listens for connection requests, until timing out or killed. Upon receipt |
| 475 |
of a cd connection request, the connection is established, the server and client sides exchange |
| 476 |
information necessary to perform RDMA writes and reads. |
| 477 |
|
| 478 |
OPTIONS |
| 479 |
-P=PROVIDER |
| 480 |
use PROVIDER to specify uDAPL interface using /etc/dat.conf (default OpenIB-cma) |
| 481 |
|
| 482 |
-b=BUFFER_SIZE |
| 483 |
use buffer size BUFFER_SIZE for RDMA(default 64) |
| 484 |
|
| 485 |
-B=BURST_COUNT |
| 486 |
use busrt count BURST_COUNT for interations (default 10) |
| 487 |
|
| 488 |
-v, verbose output(default off) |
| 489 |
|
| 490 |
-c, use consumer notification events (default off) |
| 491 |
|
| 492 |
-p, use polling (default wait for event) |
| 493 |
|
| 494 |
-d, delay in seconds before close (default off) |
| 495 |
|
| 496 |
-s, run as server (default - run as server) |
| 497 |
|
| 498 |
-h=HOSTNAME |
| 499 |
use HOSTNAME to specify server hostname or IP address (default - none) |
| 500 |
|
| 501 |
EXAMPLES |
| 502 |
dtest -P OpenIB-cma -v -s |
| 503 |
Starts a server process with debug verbosity using provider OpenIB-cma. |
| 504 |
|
| 505 |
dtest -P OpenIB-cma -h server1-ib0 |
| 506 |
|
| 507 |
Starts a client process, using OpenIB-cma provider to connect to hostname server1-ib0. |
| 508 |
|
| 509 |
SEE ALSO |
| 510 |
dapltest(1) |
| 511 |
|
| 512 |
AUTHORS |
| 513 |
Arlin Davis |
| 514 |
<[email protected]> |
| 515 |
|
| 516 |
BUGS |
| 517 |
|
| 518 |
/dapl/test/dapltest/ |
| 519 |
|
| 520 |
NAME |
| 521 |
dapltest - test for the Direct Access Programming Library (DAPL) |
| 522 |
|
| 523 |
DESCRIPTION |
| 524 |
Dapltest is a set of tests developed to exercise, characterize, and verify the DAPL interfaces |
| 525 |
during development and porting. At least two instantiations of the test must be run. One acts as |
| 526 |
the server, fielding requests and spawning server-side test threads as needed. Other client invo- |
| 527 |
cations connect to the server and issue test requests. The server side of the test, once invoked, |
| 528 |
listens continuously for client connection requests, until quit or killed. Upon receipt of a con- |
| 529 |
nection request, the connection is established, the server and client sides swap version numbers |
| 530 |
to verify that they are able to communicate, and the client sends the test request to the server. |
| 531 |
If the version numbers match, and the test request is well-formed, the server spawns the threads |
| 532 |
needed to run the test before awaiting further connections. |
| 533 |
|
| 534 |
USAGE |
| 535 |
dapltest [ -f script_file_name ] [ -T S|Q|T|P|L ] [ -D device_name ] [ -d ] [ -R HT|LL|EC|PM|BE ] |
| 536 |
|
| 537 |
With no arguments, dapltest runs as a server using default values, and loops accepting requests |
| 538 |
from clients. |
| 539 |
|
| 540 |
The -f option allows all arguments to be placed in a file, to ease test automation. |
| 541 |
|
| 542 |
The following arguments are common to all tests: |
| 543 |
|
| 544 |
[ -T S|Q|T|P|L ] |
| 545 |
Test function to be performed: |
| 546 |
|
| 547 |
S - server loop |
| 548 |
|
| 549 |
Q - quit, client requests that server wait for any outstanding tests to complete, then |
| 550 |
clean up and exit |
| 551 |
|
| 552 |
T - transaction test, transfers data between client and server |
| 553 |
|
| 554 |
P - performance test, times DTO operations |
| 555 |
|
| 556 |
L - limit test, exhausts various resources, runs in client w/o server interaction |
| 557 |
Default: S |
| 558 |
|
| 559 |
[ -D device_name ] |
| 560 |
Specifies the interface adapter name as documented in the /etc/dat.conf static configura- |
| 561 |
tion file. This name corresponds to the provider library to open. Default: none |
| 562 |
|
| 563 |
[ -d ] Enables extra debug verbosity, primarily tracing of the various DAPL operations as they |
| 564 |
progress. Repeating this parameter increases debug spew. Errors encountered result in the |
| 565 |
test spewing some explanatory text and stopping; this flag provides more detail about what |
| 566 |
lead up to the error. Default: zero |
| 567 |
|
| 568 |
[ -R BE ] |
| 569 |
Indicate the quality of service (QoS) desired. Choices are: |
| 570 |
|
| 571 |
HT - high throughput |
| 572 |
|
| 573 |
LL - low latency |
| 574 |
|
| 575 |
EC - economy (neither HT nor LL) |
| 576 |
|
| 577 |
PM - premium |
| 578 |
|
| 579 |
BE - best effort Default: BE |
| 580 |
|
| 581 |
Usage - Quit test client |
| 582 |
|
| 583 |
dapltest [Common_Args] [ -s server_name ] |
| 584 |
|
| 585 |
Quit testing (-T Q) connects to the server to ask it to clean up and |
| 586 |
exit (after it waits for any outstanding test runs to complete). |
| 587 |
In addition to being more polite than simply killing the server, |
| 588 |
this test exercises the DAPL object teardown code paths. |
| 589 |
There is only one argument other than those supported by all tests: |
| 590 |
|
| 591 |
-s server_name Specifies the name of the server interface. |
| 592 |
No default. |
| 593 |
|
| 594 |
Usage - Transaction test client |
| 595 |
|
| 596 |
dapltest [Common_Args] [ -s server_name ] |
| 597 |
[ -t threads ] [ -w endpoints ] [ -i iterations ] [ -Q ] |
| 598 |
[ -V ] [ -P ] OPclient OPserver [ op3, |
| 599 |
|
| 600 |
Transaction testing (-T T) transfers a variable amount of data between |
| 601 |
client and server. The data transfer can be described as a sequence of |
| 602 |
individual operations; that entire sequence is transferred ’iterations’ |
| 603 |
times by each thread over all of its endpoint(s). |
| 604 |
|
| 605 |
The following parameters determine the behavior of the transaction test: |
| 606 |
|
| 607 |
-s server_name Specifies the name or IP address of the server interface. |
| 608 |
No default. |
| 609 |
|
| 610 |
[ -t threads ] Specify the number of threads to be used. |
| 611 |
Default: 1 |
| 612 |
|
| 613 |
[ -w endpoints ] Specify the number of connected endpoints per thread. |
| 614 |
Default: 1 |
| 615 |
|
| 616 |
[ -i iterations ] Specify the number of times the entire sequence |
| 617 |
of data transfers will be made over each endpoint. |
| 618 |
Default: 1000 |
| 619 |
|
| 620 |
[ -Q ] Funnel completion events into a CNO. |
| 621 |
Default: use EVDs |
| 622 |
|
| 623 |
[ -V ] Validate the data being transferred. |
| 624 |
Default: ignore the data |
| 625 |
|
| 626 |
[ -P ] Turn on DTO completion polling |
| 627 |
Default: off |
| 628 |
|
| 629 |
OP1 OP2 [ OP3, ... ] |
| 630 |
A single transaction (OPx) consists of: |
| 631 |
|
| 632 |
server|client Indicates who initiates the |
| 633 |
data transfer. |
| 634 |
|
| 635 |
SR|RR|RW Indicates the type of transfer: |
| 636 |
SR send/recv |
| 637 |
RR RDMA read |
| 638 |
RW RDMA write |
| 639 |
Defaults: none |
| 640 |
|
| 641 |
[ seg_size [ num_segs ] ] |
| 642 |
: |
| 643 |
|
| 644 |
Indicates the amount and format |
| 645 |
of the data to be transferred. |
| 646 |
Default: 4096 1 |
| 647 |
(i.e., 1 4KB buffer) |
| 648 |
|
| 649 |
[ -f ] For SR transfers only, indicates |
| 650 |
that a client’s send transfer |
| 651 |
completion should be reaped when |
| 652 |
the next recv completion is reaped. |
| 653 |
Sends and receives must be paired |
| 654 |
(one client, one server, and in that |
| 655 |
order) for this option to be used. |
| 656 |
Restrictions: |
| 657 |
|
| 658 |
Due to the flow control algorithm used by the transaction test, there |
| 659 |
must be at least one SR OP for both the client and the server. |
| 660 |
|
| 661 |
Requesting data validation (-V) causes the test to automatically append |
| 662 |
three OPs to those specified. These additional operations provide |
| 663 |
synchronization points during each iteration, at which all user-specified |
| 664 |
transaction buffers are checked. These three appended operations satisfy |
| 665 |
the "one SR in each direction" requirement. |
| 666 |
|
| 667 |
The transaction OP list is printed out if -d is supplied. |
| 668 |
|
| 669 |
Usage - Performance test client |
| 670 |
|
| 671 |
dapltest [Common_Args] -s server_name [ -m p|b ] |
| 672 |
[ -i iterations ] [ -p pipeline ] OP |
| 673 |
|
| 674 |
Performance testing (-T P) times the transfer of an operation. |
| 675 |
The operation is posted ’iterations’ times. |
| 676 |
|
| 677 |
The following parameters determine the behavior of the transaction test: |
| 678 |
|
| 679 |
-s server_name Specifies the name or IP address of the server interface. |
| 680 |
No default. |
| 681 |
|
| 682 |
-m b|p Used to choose either blocking (b) or polling (p) |
| 683 |
Default: blocking (b) |
| 684 |
[ -i iterations ] Specify the number of times the entire sequence |
| 685 |
of data transfers will be made over each endpoint. |
| 686 |
Default: 1000 |
| 687 |
|
| 688 |
[ -p pipeline ] Specify the pipline length, valid arguments are in |
| 689 |
the range [0,MAX_SEND_DTOS]. If a value greater than |
| 690 |
MAX_SEND_DTOS is requested the value will be |
| 691 |
adjusted down to MAX_SEND_DTOS. |
| 692 |
Default: MAX_SEND_DTOS |
| 693 |
|
| 694 |
OP Specifies the operation as follow: |
| 695 |
|
| 696 |
RR|RW Indicates the type of transfer: |
| 697 |
RR RDMA read |
| 698 |
RW RDMA write |
| 699 |
Defaults: none |
| 700 |
|
| 701 |
[ seg_size [ num_segs ] ] |
| 702 |
Indicates the amount and format |
| 703 |
of the data to be transferred. |
| 704 |
Default: 4096 1 |
| 705 |
(i.e., 1 4KB buffer) |
| 706 |
Usage - Limit test client |
| 707 |
|
| 708 |
Limit testing (-T L) neither requires nor connects to any server |
| 709 |
instance. The client runs one or more tests which attempt to |
| 710 |
exhaust various resources to determine DAPL limits and exercise |
| 711 |
DAPL error paths. If no arguments are given, all tests are run. |
| 712 |
|
| 713 |
Limit testing creates the sequence of DAT objects needed to |
| 714 |
move data back and forth, attempting to find the limits supported |
| 715 |
for the DAPL object requested. For example, if the LMR creation |
| 716 |
limit is being examined, the test will create a set of |
| 717 |
{IA, PZ, CNO, EVD, EP} before trying to run dat_lmr_create() to |
| 718 |
failure using that set of DAPL objects. The ’width’ parameter |
| 719 |
can be used to control how many of these parallel DAPL object |
| 720 |
sets are created before beating upon the requested constructor. |
| 721 |
Use of -m limits the number of dat_*_create() calls that will |
| 722 |
be attempted, which can be helpful if the DAPL in use supports |
| 723 |
essentailly unlimited numbers of some objects. |
| 724 |
The limit test arguments are: |
| 725 |
|
| 726 |
[ -m maximum ] Specify the maximum number of dapl_*_create() |
| 727 |
attempts. |
| 728 |
Default: run to object creation failure |
| 729 |
|
| 730 |
[ -w width ] Specify the number of DAPL object sets to |
| 731 |
create while initializing. |
| 732 |
Default: 1 |
| 733 |
|
| 734 |
[ limit_ia ] Attempt to exhaust dat_ia_open() |
| 735 |
|
| 736 |
[ limit_pz ] Attempt to exhaust dat_pz_create() |
| 737 |
|
| 738 |
[ limit_cno ] Attempt to exhaust dat_cno_create() |
| 739 |
|
| 740 |
[ limit_evd ] Attempt to exhaust dat_evd_create() |
| 741 |
|
| 742 |
[ limit_ep ] Attempt to exhaust dat_ep_create() |
| 743 |
|
| 744 |
[ limit_rsp ] Attempt to exhaust dat_rsp_create() |
| 745 |
|
| 746 |
[ limit_psp ] Attempt to exhaust dat_psp_create() |
| 747 |
|
| 748 |
[ limit_lmr ] Attempt to exhaust dat_lmr_create(4KB) |
| 749 |
|
| 750 |
[ limit_rpost ] Attempt to exhaust dat_ep_post_recv(4KB) |
| 751 |
|
| 752 |
[ limit_size_lmr ] Probe maximum size dat_lmr_create() |
| 753 |
|
| 754 |
Default: run all tests |
| 755 |
EXAMPLES |
| 756 |
dapltest -T S -d -D OpenIB-cma |
| 757 |
|
| 758 |
Starts a server process with debug verbosity. |
| 759 |
|
| 760 |
dapltest -T T -d -s host1-ib0 -D OpenIB-cma -i 100 client SR 4096 2 server SR 4096 2 |
| 761 |
|
| 762 |
Runs a transaction test, with both sides |
| 763 |
sending one buffer with two 4KB segments, |
| 764 |
one hundred times. |
| 765 |
|
| 766 |
dapltest -T P -d -s host1-ib0 -D OpenIB-cma -i 100 SR 4096 2 |
| 767 |
|
| 768 |
Runs a performance test, with the client |
| 769 |
sending one buffer with two 4KB segments, |
| 770 |
one hundred times. |
| 771 |
|
| 772 |
dapltest -T Q -s host1-ib0 -D OpenIB-cma |
| 773 |
|
| 774 |
Asks the server to clean up and exit. |
| 775 |
|
| 776 |
dapltest -T L -D OpenIB-cma -d -w 16 -m 1000 |
| 777 |
|
| 778 |
Runs all of the limit tests, setting up |
| 779 |
16 complete sets of DAPL objects, and |
| 780 |
creating at most a thousand instances |
| 781 |
when trying to exhaust resources. |
| 782 |
|
| 783 |
dapltest -T T -V -d -t 2 -w 4 -i 55555 -s linux3 -D OpenIB-cma client RW 4096 1 server RW 2048 4 |
| 784 |
client SR 1024 4 server SR 4096 2 client SR 1024 3 -f server SR 2048 1 -f |
| 785 |
|
| 786 |
Runs a more complicated transaction test, |
| 787 |
with two thread using four EPs each, |
| 788 |
sending a more complicated buffer pattern |
| 789 |
for a larger number of iterations, |
| 790 |
validating the data received. |
| 791 |
|
| 792 |
============================= |
| 793 |
9.0 Summary of Fixes/Changes: |
| 794 |
============================= |
| 795 |
|
| 796 |
Release 2.1.7 (OFED 3.18-1 GA) |
| 797 |
dtest: add -a -i options, all data sizes, incremental size |
| 798 |
dapl: Fix segfault while freeing qp |
| 799 |
mpxyd: add P2P inline support for data size <= 96 bytes |
| 800 |
dtest: change rdma_write_ping_pong so client is always last receiver |
| 801 |
ucm: add DAPL_NETWORK_PROCESS_NUM option for total ranks |
| 802 |
ucm: fca create group incorrectly using IB addr instead of socket address. |
| 803 |
ucm: fca_comm_destroy called with NULL |
| 804 |
dtest: add -W option for rdma write pinpong, similiar to ib_write_lat |
| 805 |
docs: update release notes for collective build |
| 806 |
mpxyd: reduce log level for rcv message flush |
| 807 |
dapltest: dapltest with no argument not working in ppc64 arch |
| 808 |
|
| 809 |
Release 2.1.6 (OFED 3.18-1) |
| 810 |
ucm: add cluster size environments to adjust CM timers |
| 811 |
mpxyd: proxy_in data transfers can improperly start before RTU received |
| 812 |
mcm: forward open/query for MFO devices in query only mode |
| 813 |
mpxyd: byte swap incorrect on WRC wr_len |
| 814 |
dtest: remove ERR message from flush QP function |
| 815 |
dapltest: Quit command with "-n port" number will core dump |
| 816 |
config: update dat.conf for MFO qib devices, 2 adapters/ports |
| 817 |
mpxyd: add MFO support on proxy side |
| 818 |
mcm: add MFO proxy commands, device, and CM support |
| 819 |
mcm: add MFO support to openib_common code base |
| 820 |
mcm: add full offload (MFO) mode to provider to support qib on MIC |
| 821 |
dtest: pre-allocated buffer too small for RMR, DTO ops timeout |
| 822 |
mpxyd: fix buffer initialization when no-inline support is active |
| 823 |
mpxyd: reduce log level on qp_flush to CM level |
| 824 |
mcm: intra-node proxy missing LID setup on rejects |
| 825 |
mcm: add intra-node support via ibscif device and mcm provider |
| 826 |
mcm: provide MIC address info with proxy device open |
| 827 |
mcm: add device info to non-debug log |
| 828 |
common: add DAPL_DTO_TYPE_EXTENSION_IMM for rdma_write_imm DTO type checking |
| 829 |
mpxyd: fix up some of the PI logging |
| 830 |
dtest: modify rdma_write_with_msg to support uni-direction streaming |
| 831 |
mcm,mpxyd: fix dreq processing to defer QP flush when proxy WRs still pending |
| 832 |
mpxyd: update byte_len and comp_cnt for PO to remote HST communications |
| 833 |
mcm: bug fixes for non-inline devices |
| 834 |
mcm: return CM_rej with CM_req_in errors |
| 835 |
mpxyd,mcm: RDMA write with immed data not signaled on request side |
| 836 |
mcm: add WC opcode and wc_flags in debug log message |
| 837 |
mpxyd: set options bug fix for mcm_ib_inline |
| 838 |
Update release notes with latest CM times |
| 839 |
|
| 840 |
Release 2.1.5 (OFED 3.18 RC3) |
| 841 |
update release notes, readme |
| 842 |
dat.conf: update comments regarding versions |
| 843 |
dtest: add logging of provider private data size with -v |
| 844 |
scm: remove use of msg.resv field for process id logging |
| 845 |
cma: report correct CM req private data size on query |
| 846 |
mpxyd: memset ib_wr structure before post_send on WC and WR requests |
| 847 |
mcm: add HST side provider support for device without inline data capability |
| 848 |
ucm: CM changes for UD extended port space and indexer |
| 849 |
ucm: add device support for new port space hash table |
| 850 |
ucm: allocate/free AH hash table for UD endpoint types |
| 851 |
ucm: check for AH caching when destroying via UD extension |
| 852 |
ucm: optimizations for large scale UD communication management |
| 853 |
mpxyd: use wr opcode instead of wc opcode to support logging on error cases |
| 854 |
mcm: HST->MXS mode, using RDMA_WRITE_WITH_IMM, fails with dtest -w |
| 855 |
dapl: aarch64 support for linux |
| 856 |
dapltest: add scripts to dist, set default device to IPoIB |
| 857 |
mpxyd: add wc_flags to proxy work completions |
| 858 |
|
| 859 |
Release 2.1.4 (OFED 3.18 RC1) |
| 860 |
mpxyd: fix typo in configuration file |
| 861 |
cma: RR attributes moved to common ib_cm struct |
| 862 |
mpxyd: tx thread incorrectly sleeps with negative pi_rw_cnt value |
| 863 |
dat.conf: add entries for True Scale qib device |
| 864 |
mpxyd: add support for devices without inline data support |
| 865 |
ucm: long disconnect times with many-to-one applications |
| 866 |
openib: add inline data support check during device open |
| 867 |
cleanup ib/cm attribute management across openib providers |
| 868 |
dapltest: fix -Werror=format-security issue with printf |
| 869 |
Release 2.1.3 (targeting OFED 3.18) |
| 870 |
dapl: mpxyd service changes to support multi-thread single-core option |
| 871 |
dapl: add rdma_write_imm and write only option to dtest |
| 872 |
ucm: add time wait override capability for CM services |
| 873 |
common: dapl_ep_free must serialize CM object destroy |
| 874 |
dtestx: allow scale up to 1000 EP's |
| 875 |
ucm: RTU not retransmitted in TIMEWAIT state |
| 876 |
mpxyd: increase max open files for service |
| 877 |
mpxyd: DTO completion ERR: status 12, op RDMA_WRITE running MPI alltoall test |
| 878 |
mcm: HST->MXS mode incorrectly signals multiple fragments per WR |
| 879 |
mcm: add segmentation to HST->MXS mode for improved performance |
| 880 |
mpxyd: set global seg_sz to 128KB for proxy data service |
| 881 |
openib: add port_num to provider named attributes |
| 882 |
mcm: provide CPU family/model attribute on both host and mic sides |
| 883 |
dtestx: update IB extension example test with new v2.0.9 features |
| 884 |
dtest: add dtestsrq for SRQ example and provider testing |
| 885 |
common: add srq support for openib verbs providers |
| 886 |
openib: add IB UD cm_free/ah_free extension support in UCM provider |
| 887 |
openib: add new TIMEWAIT state for CM |
| 888 |
extension: add IB UD extensions to reduce provider CM and AH memory footprint |
| 889 |
mpxyd/mcm: add provider specific attribute DAT_IB_PROXY_VERSION |
| 890 |
mpxyd: log warning if running in COMPAT mode |
| 891 |
add provider and proxy support for GUID across platform |
| 892 |
common: return appropriate handles with affiliated EP and EVD async events |
| 893 |
|
| 894 |
Release 2.1.2 (OFED 3.12-1) |
| 895 |
mpxyd: add global routing support for proxy connections |
| 896 |
mcm: only call mix_get_attr if running on MIC |
| 897 |
openib: modify check for link_layer to handle unspecified |
| 898 |
dapl: add support for the s390x platform |
| 899 |
dtest server exchange connection info with client |
| 900 |
mpxyd: 2 MICs in same numa_node will overlap CPU affinity, don't reset base |
| 901 |
mcm: implement proxy mix_prov_attr function, add fields CPU model and family |
| 902 |
mpxyd: tx thread may not be signaled on small segment writes |
| 903 |
|
| 904 |
Release 2.1.1 (OFED 3.12-1 RC1) |
| 905 |
common: add provider name to log messages |
| 906 |
mpxyd: log warning message if numa_node invalid include debuginfo with build |
| 907 |
build: include debuginfo with build |
| 908 |
mpxyd: tx thread doesn't sleep during no pending IO state |
| 909 |
mpxyd: change MIC cpu_mask to per numa node instead of adapter |
| 910 |
mpxyd: set to MXS mode if device numa_node is invalid (-1) |
| 911 |
mpxyd: MXS based alltoall benchmark hangs or returns post_send timeout |
| 912 |
mpxyd: add IO profile capabilities to help debug alltoall stall cases |
| 913 |
mpxyd: retry stalled inline post_send, init m_idx only when signaled |
| 914 |
|
| 915 |
Release 2.1.0 (OFED 3.12-1, MIC support added) |
| 916 |
build: add missing NEWS file |
| 917 |
update autogen.sh |
| 918 |
add MCM provider and MPXYD service to build |
| 919 |
mpxyd: service startup script and configuration file |
| 920 |
add readme for MCM provider and MPXYD service |
| 921 |
update Copyright dates |
| 922 |
add new MIC RDMA proxy service daemon (MPXYD) |
| 923 |
add new dapl MIC provider (MCM) to support MIC RDMA proxy services |
| 924 |
MCM: new MIC provider and proxy service definitions |
| 925 |
cleanup build warnings |
| 926 |
common: add CQ,QP,MR abstractions for new MIC provider and data proxy service |
| 927 |
openib: cleanup, use inet_ntop for GIDs, remove some logs, destroy pipes on release |
| 928 |
common: new dapls_evd_cqe_to_event call, cqe to event |
| 929 |
common: init ring_buffer, assign hd/tl pos in range |
| 930 |
allow log level changes during device open |
| 931 |
ucm: fix cm rbuf setup, include grh pad on initialization |
| 932 |
ucm: remove duplicate async_event code, use common async event call |
| 933 |
new lightweight open_query/close_query IB extension for fast attribute query |
| 934 |
dtestcm: add more detailed debug during disconnect phase |
| 935 |
cma: long delays when opening cma provider with no IPoIB configured |
| 936 |
common: new debug levels for low system memory, IA stats, and package info |
| 937 |
build: remove library check for mverbs with --enable-fca |
| 938 |
IB extension: segfault in create collective group with non-vector type IA handle" |
| 939 |
build: change configure help to correctly state collective default=none |
| 940 |
|
| 941 |
Release 2.0.42 fixes (OFED 3.12 GA) |
| 942 |
dapltest: increase DTO evd size to prevent CQ overflow on limit_rpost test |
| 943 |
dapltest: RSP limit test fails. Creation of reserved SP moves EP state to DAT_EP_STATE_RESERVED in error cases. |
| 944 |
dapl: fix string bug in dapls_dto_op_str |
| 945 |
|
| 946 |
Release 2.0.41 fixes (OFED 3.12 RC1) |
| 947 |
dapltest: change server port, from 45278 to 62000, out of registered IANA range |
| 948 |
dat: lower log level on load errors of provider library |
| 949 |
dat: dat_ia_open should close provider after failure |
| 950 |
dapltest: set default limit max to 1000 |
| 951 |
openib: add new provider specific attributes |
| 952 |
dapltest: update scripts for regression testing purposes |
| 953 |
dapltest: Add final send/recv "sync" for transaction tests. |
| 954 |
|
| 955 |
Release 2.0.40 fixes (OFED 3.12) |
| 956 |
dist: ib collective extension include files missing |
| 957 |
dapltest: the quit command is missing changes for -n option |
| 958 |
dat.conf: remove v1, add Mellanox Connect-IB and Intel Xeon Phi MIC |
| 959 |
NULL undefined on Fedora, incorrectly using kernel stddef.h |
| 960 |
|
| 961 |
Release 2.0.39 fixes (OFED 3.5-2 GA) |
| 962 |
dapltest: fix endian swap issue with performance test |
| 963 |
scm: getifaddrs modfications for better out of the box experience |
| 964 |
ucm, scm: UD mode triggers list_head assert with large scale alltoall test |
| 965 |
|
| 966 |
Release 2.0.38 |
| 967 |
dapltest: add -n parameter to override default server port number (45278) |
| 968 |
ucm,scm: UD mode creates many CR objects per EP that needs cleaned up |
| 969 |
cma: add DAPL_CM_TOS environment variable to enable passing a TOS to the RDMA CM |
| 970 |
|
| 971 |
Release 2.0.37 |
| 972 |
common: add support for ia name during dat_ia_query |
| 973 |
common: dapl_os_atomic_inc/dec() not working as expected on ppc64 machines. |
| 974 |
dapltest: ppc64 endian issue with exchanged mem handle and address |
| 975 |
|
| 976 |
Release 2.0.36 |
| 977 |
scm: increase ACK timeout to 20 for a default value to match other providers. |
| 978 |
common: allow qp modify in init state |
| 979 |
common: check for valid states during ep posting |
| 980 |
dat.conf: keep list of providers in order for backward compatibility |
| 981 |
ucm: record and silently drop a duplicate reject CM message |
| 982 |
windows: new version of getlocalipaddr not portable |
| 983 |
dapltest: DFLT_QLEN is defined in multiple tests |
| 984 |
|
| 985 |
Release 2.0.35 |
| 986 |
config/build: remove post/postun hacking used to modify dat.conf |
| 987 |
config: clean up help option displays with ext-type options |
| 988 |
windows: Provide auto-detect between RoCE and Infiniband for Windows. |
| 989 |
ucm: update UD cm provider to support new CM stat and error counters |
| 990 |
scm: update socket cm provider to support new CM stat and error counters |
| 991 |
commom: add cm, link, and diag event counters in IB extended builds |
| 992 |
scm: use ioctl SIOCIFCONF to get complete list of configured netdev interfaces |
| 993 |
ucm: UD send failures at scale, ucm_send ERR: get_smsg(hd=149,tl=150) |
| 994 |
scm: fix retry count on connection pending timeout |
| 995 |
ucm: cleanup debug message, ntohl on p_size is incorrect |
| 996 |
cma, scm, ucm: allow EP (QP) creation without EVD (CQ) |
| 997 |
common: add DAPL_DBG_TYPE_CM_STATS (0x40000) to debug log options |
| 998 |
common: dapls_ep_flush_cq will segfault when no CQ is attached to EP |
| 999 |
common: ep_create should allow max_request_iov attribute setting of zero |
| 1000 |
common: add check for NULL handle on ext calls, SRQ free, and helper functions |
| 1001 |
common: add missing sub-types to dat_strerror() |
| 1002 |
common: extended CR event processing missing rejects on errors |
| 1003 |
ucm: incorrectly sends user reject during CR callback errors |
| 1004 |
common: change dbg level on CR callback if not listening on SP |
| 1005 |
scm: incorrectly sends user reject during CR callback errors |
| 1006 |
dat: add check for NULL handle on IA calls |
| 1007 |
cma,scm,ucm: extra reference on EP, with RSP, causes dat_ep_free() to hang |
| 1008 |
common: RSP service points incorrectly freed during CR callback |
| 1009 |
common: clean up dat_rsp_create log message |
| 1010 |
common: cleanup debug message on EVD overflows |
| 1011 |
scm: return correct event error code when remote host refuses requests |
| 1012 |
dapltest: server CR EVD is too small for multi-client configurations. |
| 1013 |
Common: CR EVD overflow causes segfault. |
| 1014 |
|
| 1015 |
Release 2.0.34 |
| 1016 |
scm: change debug message level for listen/bind errors |
| 1017 |
common: increase default IB ack timer from 16 to 20 |
| 1018 |
common: remote ia address null pointer creates seg fault |
| 1019 |
common: posting events on full queue returns wrong error code |
| 1020 |
common: dat_ep_modify seg faults with null ep_param ptr |
| 1021 |
common: dat_evd_free seg faults with resized software EVD |
| 1022 |
common: remove assert for incorrect events during cm_request |
| 1023 |
dat: dat_cno_query with NULL cno_handle causes segmentation fault |
| 1024 |
scm: dat_psp_create returns wrong error code on bind/listen failure |
| 1025 |
scm: socket connect request count is reset improperly on retry |
| 1026 |
scm: when hostname has loopback addr assigned, default to eth0 instead of failing |
| 1027 |
scm: add port number to error log during hca_open failures |
| 1028 |
common: query calls return incorrect IA handle to consumer |
| 1029 |
common: srq create asserts with !dapl_llist_is_empty(head) failed |
| 1030 |
|
| 1031 |
Release 2.0.33 |
| 1032 |
scm,ucm: fix compatibility issues and set minimum protocol support |
| 1033 |
build: link librdmacm dependency to ib_acm usage for ucm and scm providers |
| 1034 |
build: add selective enable/disable-xxx build switch for each provider |
| 1035 |
build: add extended header files to EXTRA_DIST and fix missing backslash |
| 1036 |
build: set IB extended coll-type to none by default |
| 1037 |
common: change errno mapping of EINVAL to DAT_INVALID_PARAMETER |
| 1038 |
build: add IB collective and FCA provider to dapl build package as an option |
| 1039 |
common: add new dapls_evd_post_event_ext call for extended events |
| 1040 |
ucm: add support for IB collective providers |
| 1041 |
scm: add support for IB collective providers |
| 1042 |
cma: add support for IB collective providers |
| 1043 |
common: add supported collective types in named attributes for query |
| 1044 |
common: add collective call mappings via standard dapli_post_ext() |
| 1045 |
common: new debug bitmask definition for extension logging |
| 1046 |
common: new IB collective provider for Mellanox Fabric Collective Agent |
| 1047 |
dat: add definitions for MPI offloaded collectives in IB transport extensions |
| 1048 |
common: cleanup debug messages when building with ibacm feature |
| 1049 |
|
| 1050 |
Release 2.0.32 fixes (OFED 1.5.3 GA): |
| 1051 |
|
| 1052 |
cma: reduce output log level in disconnect from WARN to CM_WARN |
| 1053 |
ucm: delay freeing of active side UD cm object in case RTU is dropped |
| 1054 |
ucm: cm object needs to be on work queue before req sent on wire |
| 1055 |
ucm,scm: remove use of usec_sleep delays and use events for disc and destroy |
| 1056 |
common: reduce default max inline data size because of performance anomaly |
| 1057 |
common: dapls_evd_dto_wait() dbg message should print status and not errno |
| 1058 |
ucm, scm: exchange max_qp_rd_atom and limit outstanding requests |
| 1059 |
scm: retry socket connect on ECONNREFUSED under heavy load |
| 1060 |
common: qp modify RTR using wrong ep attribute parameter for dest_rd_atomic |
| 1061 |
|
| 1062 |
Release 2.0.31 fixes (OFED 1.5.3 RC1): |
| 1063 |
|
| 1064 |
common: clean up build warning for unused variable event_ptr |
| 1065 |
scm, ucm: set RAI_NOROUTE flag with rdma_getaddrinfo() call to avoid blocking. |
| 1066 |
cma: definition for dapl_sp_remove_ep() is missing in cm.c |
| 1067 |
libdat: static provider entries created for local SR database not freed |
| 1068 |
libdat: memory leak in static registration during parsing |
| 1069 |
common: increase default IB inline send threshold to 400 |
| 1070 |
common cq: a mixup of errno and the -1 return from poll in dapls_wait_comp_channel |
| 1071 |
ucm: release UD cm objects after AH is exchanged to avoid duplicate request drops |
| 1072 |
ucm: decrease timeout retry count for disconnect requests |
| 1073 |
ucm: hold lock when sending cm_msgs to sync timer start with packet send |
| 1074 |
ucm: add debugging to include process id for better scale up debug aids |
| 1075 |
cma: disconnect can block for excessive times waiting for rdma_cm DREP timeout |
| 1076 |
ucm: configure the recv channel FD to non-blocking |
| 1077 |
windows: Missing librdmacm include path for build |
| 1078 |
debug build: only timestamp if sending to stdout to avoid performance hit |
| 1079 |
common: print out errors on free build and not just debug builds |
| 1080 |
cma: fix debug build issue |
| 1081 |
scm, ucm: MPI spawn test on oversubcribed server taking excessive time to complete |
| 1082 |
common: add high resolution time stamps and thread id to sdtout debug logs |
| 1083 |
common: modify debug in dat_evd_dequeue to reduce noise, only output on non-empty |
| 1084 |
cma: rdma_destroy_id called twice during device open bind error |
| 1085 |
common: dat_evd_dequeue (poll_cq) fails with invalid parameter after EP (qp) free |
| 1086 |
ucm: allow configuration of CM burst (signal) threshold on posting |
| 1087 |
cma: fix debug build |
| 1088 |
windows: debug version of windows does not build. |
| 1089 |
Allow DAPL out of band connection models to use ibacm to obtain path record data. |
| 1090 |
ucm: add missing map file for UCM provider |
| 1091 |
ibal: delay QP transition during disconnect phase |
| 1092 |
Revert "ibal: delay QP transition during disconnect phase" |
| 1093 |
ibal: delay QP transition during disconnect phase |
| 1094 |
common: restructure EVD processing to handle EP destruction phase |
| 1095 |
ibal: sync QP destruction and device close |
| 1096 |
ucm: remove unnecessary debug warning in async callback |
| 1097 |
|