| 1 |
|
| 2 |
NAME |
| 3 |
|
| 4 |
dapltest - test for the Direct Access Programming Library (DAPL) |
| 5 |
|
| 6 |
DESCRIPTION |
| 7 |
|
| 8 |
Dapltest is a set of tests developed to exercise, characterize, |
| 9 |
and verify the DAPL interfaces during development and porting. |
| 10 |
At least two instantiations of the test must be run. One acts |
| 11 |
as the server, fielding requests and spawning server-side test |
| 12 |
threads as needed. Other client invocations connect to the |
| 13 |
server and issue test requests. |
| 14 |
|
| 15 |
The server side of the test, once invoked, listens continuously |
| 16 |
for client connection requests, until quit or killed. Upon |
| 17 |
receipt of a connection request, the connection is established, |
| 18 |
the server and client sides swap version numbers to verify that |
| 19 |
they are able to communicate, and the client sends the test |
| 20 |
request to the server. If the version numbers match, and the |
| 21 |
test request is well-formed, the server spawns the threads |
| 22 |
needed to run the test before awaiting further connections. |
| 23 |
|
| 24 |
USAGE |
| 25 |
|
| 26 |
dapltest [ -f script_file_name ] |
| 27 |
[ -T S|Q|T|P|L ] [ -D device_name ] [ -d ] [ -R HT|LL|EC|PM|BE ] |
| 28 |
|
| 29 |
With no arguments, dapltest runs as a server using default values, |
| 30 |
and loops accepting requests from clients. The -f option allows |
| 31 |
all arguments to be placed in a file, to ease test automation. |
| 32 |
The following arguments are common to all tests: |
| 33 |
|
| 34 |
[ -T S|Q|T|P|L ] Test function to be performed: |
| 35 |
S - server loop |
| 36 |
Q - quit, client requests that server |
| 37 |
wait for any outstanding tests to |
| 38 |
complete, then clean up and exit |
| 39 |
T - transaction test, transfers data between |
| 40 |
client and server |
| 41 |
P - performance test, times DTO operations |
| 42 |
L - limit test, exhausts various resources, |
| 43 |
runs in client w/o server interaction |
| 44 |
Default: S |
| 45 |
|
| 46 |
[ -D device_name ] Specifies the name of the device (interface adapter). |
| 47 |
Default: host-specific, look for DT_MdepDeviceName |
| 48 |
in dapl_mdep.h |
| 49 |
|
| 50 |
[ -d ] Enables extra debug verbosity, primarily tracing |
| 51 |
of the various DAPL operations as they progress. |
| 52 |
Repeating this parameter increases debug spew. |
| 53 |
Errors encountered result in the test spewing some |
| 54 |
explanatory text and stopping; this flag provides |
| 55 |
more detail about what lead up to the error. |
| 56 |
Default: zero |
| 57 |
|
| 58 |
[ -R BE ] Indicate the quality of service (QoS) desired. |
| 59 |
Choices are: |
| 60 |
HT - high throughput |
| 61 |
LL - low latency |
| 62 |
EC - economy (neither HT nor LL) |
| 63 |
PM - premium |
| 64 |
BE - best effort |
| 65 |
Default: BE |
| 66 |
|
| 67 |
USAGE - Quit test client |
| 68 |
|
| 69 |
dapltest [Common_Args] [ -s server_name ] |
| 70 |
|
| 71 |
Quit testing (-T Q) connects to the server to ask it to clean up and |
| 72 |
exit (after it waits for any outstanding test runs to complete). |
| 73 |
In addition to being more polite than simply killing the server, |
| 74 |
this test exercises the DAPL object teardown code paths. |
| 75 |
There is only one argument other than those supported by all tests: |
| 76 |
|
| 77 |
-s server_name Specifies the name of the server interface. |
| 78 |
No default. |
| 79 |
|
| 80 |
|
| 81 |
USAGE - Transaction test client |
| 82 |
|
| 83 |
dapltest [Common_Args] [ -s server_name ] |
| 84 |
[ -t threads ] [ -w endpoints ] [ -i iterations ] [ -Q ] |
| 85 |
[ -V ] [ -P ] OPclient OPserver [ op3, |
| 86 |
|
| 87 |
Transaction testing (-T T) transfers a variable amount of data between |
| 88 |
client and server. The data transfer can be described as a sequence of |
| 89 |
individual operations; that entire sequence is transferred 'iterations' |
| 90 |
times by each thread over all of its endpoint(s). |
| 91 |
|
| 92 |
The following parameters determine the behavior of the transaction test: |
| 93 |
|
| 94 |
-s server_name Specifies the name of the server interface. |
| 95 |
No default. |
| 96 |
|
| 97 |
[ -t threads ] Specify the number of threads to be used. |
| 98 |
Default: 1 |
| 99 |
|
| 100 |
[ -w endpoints ] Specify the number of connected endpoints per thread. |
| 101 |
Default: 1 |
| 102 |
|
| 103 |
[ -i iterations ] Specify the number of times the entire sequence |
| 104 |
of data transfers will be made over each endpoint. |
| 105 |
Default: 1000 |
| 106 |
|
| 107 |
[ -Q ] Funnel completion events into a CNO. |
| 108 |
Default: use EVDs |
| 109 |
|
| 110 |
[ -V ] Validate the data being transferred. |
| 111 |
Default: ignore the data |
| 112 |
|
| 113 |
[ -P ] Turn on DTO completion polling |
| 114 |
Default: off |
| 115 |
|
| 116 |
OP1 OP2 [ OP3, ... ] |
| 117 |
A single transaction (OPx) consists of: |
| 118 |
|
| 119 |
server|client Indicates who initiates the |
| 120 |
data transfer. |
| 121 |
|
| 122 |
SR|RR|RW Indicates the type of transfer: |
| 123 |
SR send/recv |
| 124 |
RR RDMA read |
| 125 |
RW RDMA write |
| 126 |
Defaults: none |
| 127 |
|
| 128 |
[ seg_size [ num_segs ] ] |
| 129 |
Indicates the amount and format |
| 130 |
of the data to be transferred. |
| 131 |
Default: 4096 1 |
| 132 |
(i.e., 1 4KB buffer) |
| 133 |
|
| 134 |
[ -f ] For SR transfers only, indicates |
| 135 |
that a client's send transfer |
| 136 |
completion should be reaped when |
| 137 |
the next recv completion is reaped. |
| 138 |
Sends and receives must be paired |
| 139 |
(one client, one server, and in that |
| 140 |
order) for this option to be used. |
| 141 |
|
| 142 |
Restrictions: |
| 143 |
|
| 144 |
Due to the flow control algorithm used by the transaction test, there |
| 145 |
must be at least one SR OP for both the client and the server. |
| 146 |
|
| 147 |
Requesting data validation (-V) causes the test to automatically append |
| 148 |
three OPs to those specified. These additional operations provide |
| 149 |
synchronization points during each iteration, at which all user-specified |
| 150 |
transaction buffers are checked. These three appended operations satisfy |
| 151 |
the "one SR in each direction" requirement. |
| 152 |
|
| 153 |
The transaction OP list is printed out if -d is supplied. |
| 154 |
|
| 155 |
USAGE - Performance test client |
| 156 |
|
| 157 |
dapltest [Common_Args] -s server_name [ -m p|b ] |
| 158 |
[ -i iterations ] [ -p pipeline ] OP |
| 159 |
|
| 160 |
Performance testing (-T P) times the transfer of an operation. |
| 161 |
The operation is posted 'iterations' times. |
| 162 |
|
| 163 |
The following parameters determine the behavior of the transaction test: |
| 164 |
|
| 165 |
-s server_name Specifies the name of the server interface. |
| 166 |
No default. |
| 167 |
|
| 168 |
-m b|p Used to choose either blocking (b) or polling (p) |
| 169 |
Default: blocking (b) |
| 170 |
|
| 171 |
[ -i iterations ] Specify the number of times the entire sequence |
| 172 |
of data transfers will be made over each endpoint. |
| 173 |
Default: 1000 |
| 174 |
|
| 175 |
[ -p pipeline ] Specify the pipline length, valid arguments are in |
| 176 |
the range [0,MAX_SEND_DTOS]. If a value greater than |
| 177 |
MAX_SEND_DTOS is requested the value will be |
| 178 |
adjusted down to MAX_SEND_DTOS. |
| 179 |
Default: MAX_SEND_DTOS |
| 180 |
|
| 181 |
OP |
| 182 |
An operation consists of: |
| 183 |
|
| 184 |
RR|RW Indicates the type of transfer: |
| 185 |
RR RDMA read |
| 186 |
RW RDMA write |
| 187 |
Default: none |
| 188 |
|
| 189 |
[ seg_size [ num_segs ] ] |
| 190 |
Indicates the amount and format |
| 191 |
of the data to be transferred. |
| 192 |
Default: 4096 1 |
| 193 |
(i.e., 1 4KB buffer) |
| 194 |
|
| 195 |
USAGE - Limit test client |
| 196 |
|
| 197 |
Limit testing (-T L) neither requires nor connects to any server |
| 198 |
instance. The client runs one or more tests which attempt to |
| 199 |
exhaust various resources to determine DAPL limits and exercise |
| 200 |
DAPL error paths. If no arguments are given, all tests are run. |
| 201 |
|
| 202 |
Limit testing creates the sequence of DAT objects needed to |
| 203 |
move data back and forth, attempting to find the limits supported |
| 204 |
for the DAPL object requested. For example, if the LMR creation |
| 205 |
limit is being examined, the test will create a set of |
| 206 |
{IA, PZ, CNO, EVD, EP} before trying to run dat_lmr_create() to |
| 207 |
failure using that set of DAPL objects. The 'width' parameter |
| 208 |
can be used to control how many of these parallel DAPL object |
| 209 |
sets are created before beating upon the requested constructor. |
| 210 |
Use of -m limits the number of dat_*_create() calls that will |
| 211 |
be attempted, which can be helpful if the DAPL in use supports |
| 212 |
essentailly unlimited numbers of some objects. |
| 213 |
|
| 214 |
The limit test arguments are: |
| 215 |
|
| 216 |
[ -m maximum ] Specify the maximum number of dapl_*_create() |
| 217 |
attempts. |
| 218 |
Default: run to object creation failure |
| 219 |
|
| 220 |
[ -w width ] Specify the number of DAPL object sets to |
| 221 |
create while initializing. |
| 222 |
Default: 1 |
| 223 |
|
| 224 |
[ limit_ia ] Attempt to exhaust dat_ia_open() |
| 225 |
|
| 226 |
[ limit_pz ] Attempt to exhaust dat_pz_create() |
| 227 |
|
| 228 |
[ limit_cno ] Attempt to exhaust dat_cno_create() |
| 229 |
|
| 230 |
[ limit_evd ] Attempt to exhaust dat_evd_create() |
| 231 |
|
| 232 |
[ limit_ep ] Attempt to exhaust dat_ep_create() |
| 233 |
|
| 234 |
[ limit_rsp ] Attempt to exhaust dat_rsp_create() |
| 235 |
|
| 236 |
[ limit_psp ] Attempt to exhaust dat_psp_create() |
| 237 |
|
| 238 |
[ limit_lmr ] Attempt to exhaust dat_lmr_create(4KB) |
| 239 |
|
| 240 |
[ limit_rpost ] Attempt to exhaust dat_ep_post_recv(4KB) |
| 241 |
|
| 242 |
[ limit_size_lmr ] Probe maximum size dat_lmr_create() |
| 243 |
|
| 244 |
Default: run all tests |
| 245 |
|
| 246 |
|
| 247 |
EXAMPLES |
| 248 |
|
| 249 |
dapltest -T S -d -D jni0a |
| 250 |
|
| 251 |
Starts a server process with debug verbosity. |
| 252 |
|
| 253 |
dapltest -T T -d -s linux3 -D jni0a -i 100 \ |
| 254 |
client SR 4096 2 server SR 4096 2 |
| 255 |
|
| 256 |
Runs a transaction test, with both sides |
| 257 |
sending one buffer with two 4KB segments, |
| 258 |
one hundred times. |
| 259 |
|
| 260 |
dapltest -T P -d -s linux3 -D JniIbdd0 -i 100 SR 4096 2 |
| 261 |
|
| 262 |
Runs a performance test, with the client |
| 263 |
sending one buffer with two 4KB segments, |
| 264 |
one hundred times. |
| 265 |
|
| 266 |
dapltest -T Q -s linux3 -D jni0a |
| 267 |
|
| 268 |
Asks the server to clean up and exit. |
| 269 |
|
| 270 |
dapltest -T L -D jni0a -d -w 16 -m 1000 |
| 271 |
|
| 272 |
Runs all of the limit tests, setting up |
| 273 |
16 complete sets of DAPL objects, and |
| 274 |
creating at most a thousand instances |
| 275 |
when trying to exhaust resources. |
| 276 |
|
| 277 |
dapltest -T T -V -d -t 2 -w 4 -i 55555 -s linux3 -D jni0a \ |
| 278 |
client RW 4096 1 server RW 2048 4 \ |
| 279 |
client SR 1024 4 server SR 4096 2 \ |
| 280 |
client SR 1024 3 -f server SR 2048 1 -f |
| 281 |
|
| 282 |
Runs a more complicated transaction test, |
| 283 |
with two thread using four EPs each, |
| 284 |
sending a more complicated buffer pattern |
| 285 |
for a larger number of iterations, |
| 286 |
validating the data received. |
| 287 |
|
| 288 |
|
| 289 |
BUGS (and To Do List) |
| 290 |
|
| 291 |
Use of CNOs (-Q) is not yet supported. |
| 292 |
|
| 293 |
Further limit tests could be added. |