forked from JeffersonLab/hdrdmacp
-
Notifications
You must be signed in to change notification settings - Fork 0
/
hdRDMA.cc
423 lines (360 loc) · 12.8 KB
/
hdRDMA.cc
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
#include <hdRDMA.h>
#include <unistd.h>
#include <string.h>
#include <strings.h>
#include <iostream>
#include <atomic>
#include <chrono>
#include <fstream>
using std::cout;
using std::cerr;
using std::endl;
using namespace std::chrono;
using std::chrono::duration;
using std::chrono::duration_cast;
using std::chrono::high_resolution_clock;
extern uint64_t HDRDMA_BUFF_LEN_GB;
extern uint64_t HDRDMA_NUM_BUFF_SECTIONS;
//-------------------------------------------------------------
// hdRDMA
//
// hdRDMA constructor. This will look for IB devices and set up
// for RDMA communications on the first one it finds.
//-------------------------------------------------------------
hdRDMA::hdRDMA()
{
cout << "Looking for IB devices ..." << endl;
int num_devices = 0;
struct ibv_device **devs = ibv_get_device_list( &num_devices );
// List devices
cout << endl << "=============================================" << endl;
cout << "Found " << num_devices << " devices" << endl;
cout << "---------------------------------------------" << endl;
for(int i=0; i<num_devices; i++){
const char *transport_type = "unknown";
switch( devs[i]->transport_type ){
case IBV_TRANSPORT_IB:
transport_type = "IB";
break;
case IBV_TRANSPORT_IWARP:
transport_type = "IWARP";
break;
case IBV_EXP_TRANSPORT_SCIF:
transport_type = "SCIF";
break;
default:
transport_type = "UNKNOWN";
break;
}
// - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
// Here we want to check the lid of each device but to do so we
// must open the device and get the port attributes. We need to
// do this to determine which device is actually connected to the
// IB network since only connected ones will have lid!=0.
// We remember the last device in the list with a non-zero lid
// and use that.
uint64_t lid = 0;
// Open device
ctx = ibv_open_device(devs[i]);
int Nports = 0;
if( ctx ){
// Loop over port numbers
for( uint8_t port_num = 1; port_num<10; port_num++) { // (won't be more than 2!)
struct ibv_port_attr my_port_attr;
auto ret = ibv_query_port( ctx, port_num, &my_port_attr);
if( ret != 0 ) break;
Nports++;
if( my_port_attr.lid != 0){
lid = my_port_attr.lid;
dev = devs[i];
this->port_num = port_num;
}
}
ibv_close_device( ctx );
}
// - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
cout << " device " << i
<< " : " << devs[i]->name
<< " : " << devs[i]->dev_name
<< " : " << transport_type
<< " : " << ibv_node_type_str(devs[i]->node_type)
<< " : Num. ports=" << Nports
<< " : port num=" << port_num
// << " : GUID=" << ibv_get_device_guid(devs[i])
<< " : lid=" << lid
<< endl;
}
cout << "=============================================" << endl << endl;
// Open device
ctx = ibv_open_device(dev);
if( !ctx ){
cout << "Error opening IB device context!" << endl;
exit(-11);
}
// Get device and port attributes
int index = 0;
ibv_gid gid;
ibv_query_device( ctx, &attr);
ibv_query_port( ctx, port_num, &port_attr);
ibv_query_gid(ctx, port_num, index, &gid);
cout << "Device " << dev->name << " opened."
<< " num_comp_vectors=" << ctx->num_comp_vectors
<< endl;
// Print some of the port attributes
cout << "Port attributes:" << endl;
cout << " state: " << port_attr.state << endl;
cout << " max_mtu: " << port_attr.max_mtu << endl;
cout << " active_mtu: " << port_attr.active_mtu << endl;
cout << " port_cap_flags: " << port_attr.port_cap_flags << endl;
cout << " max_msg_sz: " << port_attr.max_msg_sz << endl;
cout << " active_width: " << (uint64_t)port_attr.active_width << endl;
cout << " active_speed: " << (uint64_t)port_attr.active_speed << endl;
cout << " phys_state: " << (uint64_t)port_attr.phys_state << endl;
cout << " link_layer: " << (uint64_t)port_attr.link_layer << endl;
// Allocate protection domain
pd = ibv_alloc_pd(ctx);
if( !pd ){
cout << "ERROR allocation protection domain!" << endl;
exit(-12);
}
// Allocate a large buffer and create a memory region pointing to it.
// We will split this one memory region among multiple receive requests
// n.b. initial tests failed on transfer for buffers larger than 1GB
uint64_t buff_len_GB = HDRDMA_BUFF_LEN_GB;
num_buff_sections = HDRDMA_NUM_BUFF_SECTIONS;
buff_section_len = (buff_len_GB*1000000000)/(uint64_t)num_buff_sections;
buff_len = num_buff_sections*buff_section_len;
buff = new uint8_t[buff_len];
if( !buff ){
cout << "ERROR: Unable to allocate buffer!" << endl;
exit(-13);
}
errno = 0;
auto access = IBV_ACCESS_LOCAL_WRITE | IBV_ACCESS_REMOTE_WRITE;
mr = ibv_reg_mr( pd, buff, buff_len, access);
if( !mr ){
cout << "ERROR: Unable to register memory region! errno=" << errno << endl;
cout << " (Please see usage statement for a possible work around)" << endl;
exit( -14 );
}
// Fill in buffers
for( uint32_t i=0; i<num_buff_sections; i++){
auto b = &buff[ i*buff_section_len ];
hdRDMAThread::bufferinfo bi = std::make_tuple( b, buff_section_len );
buffer_pool.push_back( bi );
}
cout << "Created " << buffer_pool.size() << " buffers of " << buff_section_len/1000000 << "MB (" << buff_len/1000000000 << "GB total)" << endl;
// Create thread to listen for async ibv events
new std::thread( [&](){
while( !done ){
struct ibv_async_event async_event;
auto ret = ibv_get_async_event( ctx, &async_event);
cout << "+++ RDMA async event: type=" << async_event.event_type << " ret=" << ret << endl;
ibv_ack_async_event( &async_event );
}
});
Ntransferred = 0;
t_last = high_resolution_clock::now();
}
//-------------------------------------------------------------
// ~hdRDMA
//-------------------------------------------------------------
hdRDMA::~hdRDMA()
{
// Stop all connection threads
for( auto t : threads ){
t.second->stop = true;
t.first->join();
delete t.second;
}
// Close and free everything
if( mr!=nullptr ) ibv_dereg_mr( mr );
if( buff!=nullptr ) delete[] buff;
if( pd!=nullptr ) ibv_dealloc_pd( pd );
if( ctx!=nullptr ) ibv_close_device( ctx );
if( server_sockfd ) shutdown( server_sockfd, SHUT_RDWR );
}
//-------------------------------------------------------------
// Listen
//
// Set up server and listen for connections from remote hosts
// wanting to trade RDMA connection information.
//-------------------------------------------------------------
void hdRDMA::Listen(int port)
{
// Create socket, bind it and put it into the listening state.
struct sockaddr_in addr;
bzero( &addr, sizeof(addr) );
addr.sin_family = AF_INET;
addr.sin_addr.s_addr = htonl(INADDR_ANY);
addr.sin_port = htons( port );
server_sockfd = socket(AF_INET, SOCK_STREAM, 0);
auto ret = bind( server_sockfd, (struct sockaddr*)&addr, sizeof(addr) );
if( ret != 0 ){
cout << "ERROR: binding server socket!" << endl;
exit(-2);
}
listen(server_sockfd, 5);
// Create separate thread to accept socket connections so we don't block
std::atomic<bool> thread_started(false);
server_thread = new std::thread([&](){
// Loop forever accepting connections
cout << "Listening for connections on port ... " << port << endl;
thread_started = true;
while( !done ){
int peer_sockfd = 0;
struct sockaddr_in peer_addr;
socklen_t peer_addr_len = sizeof(struct sockaddr_in);
peer_sockfd = accept(server_sockfd, (struct sockaddr *)&peer_addr, &peer_addr_len);
if( peer_sockfd > 0 ){
// cout << "Connection from " << inet_ntoa(peer_addr.sin_addr) << endl;
// Create a new thread to handle this connection
auto hdthr = new hdRDMAThread( this );
auto thr = new std::thread( &hdRDMAThread::ThreadRun, hdthr, peer_sockfd );
std::lock_guard<std::mutex> lck( threads_mtx );
threads[ thr ] = hdthr;
Nconnections++;
}else{
cout << "Failed connection! errno=" << errno <<endl;
//break;
}
} // !done
cout << "TCP server stopped." << endl;
});
// Wait for thread to start up so it's listening message gets printed
// before rest of program continues.
while(!thread_started) std::this_thread::yield();
}
//-------------------------------------------------------------
// StopListening
//-------------------------------------------------------------
void hdRDMA::StopListening(void)
{
if( server_thread ){
cout << "Waiting for server to finish ..." << endl;
done = true;
server_thread->join();
delete server_thread;
server_thread = nullptr;
if( server_sockfd ) close( server_sockfd );
server_sockfd = 0;
}else{
cout << "Server not running." <<endl;
}
}
//-------------------------------------------------------------
// Connect
//-------------------------------------------------------------
void hdRDMA::Connect(std::string host, int port)
{
// Get IP address based on server hostname
struct addrinfo hints;
struct addrinfo *result;
char addrstr[100];
void *ptr = nullptr;
memset(&hints, 0, sizeof(struct addrinfo));
hints.ai_socktype = SOCK_STREAM;
hints.ai_family = AF_UNSPEC;
hints.ai_flags |= AI_CANONNAME;
auto ret = getaddrinfo(host.c_str(), NULL, &hints, &result);
while( result ){
inet_ntop( result->ai_family, result->ai_addr->sa_data, addrstr, 100);
switch( result->ai_family ){
case AF_INET:
ptr = &((struct sockaddr_in *)result->ai_addr)->sin_addr;
break;
case AF_INET6:
ptr = &((struct sockaddr_in6 *)result->ai_addr)->sin6_addr;
break;
}
inet_ntop( result->ai_family, ptr, addrstr, 100 );
cout << "IP address: " << addrstr << " (" << result->ai_canonname << ")" << endl;
result = result->ai_next;
}
// Create socket and connect it to remote host
struct sockaddr_in addr;
bzero( &addr, sizeof(addr) );
addr.sin_family = AF_INET;
addr.sin_addr.s_addr = inet_addr( addrstr );
addr.sin_port = htons( port );
int sockfd = socket(AF_INET, SOCK_STREAM, 0);
ret = connect( sockfd, (struct sockaddr*)&addr, sizeof(addr) );
if( ret != 0 ){
cout << "ERROR: connecting to server: " << host << " (" << inet_ntoa(addr.sin_addr) << ")" << endl;
exit(-3);
}else{
cout << "Connected to " << host << ":" << port << endl;
}
// Create an hdRDMAThread object to handle the RDMA connection details.
// (we won't actually run it in a separate thread.)
hdthr_client = new hdRDMAThread( this );
hdthr_client->ClientConnect( sockfd );
}
//-------------------------------------------------------------
// GetNpeers
//-------------------------------------------------------------
uint32_t hdRDMA::GetNpeers(void)
{
return threads.size();
}
//-------------------------------------------------------------
// GetBuffers
//-------------------------------------------------------------
void hdRDMA::GetBuffers( std::vector<hdRDMAThread::bufferinfo> &buffers, int Nrequested )
{
std::lock_guard<std::mutex> grd( buffer_pool_mutex );
//cout << "buffer_pool.size()="<<buffer_pool.size() << " Nrequested=" << Nrequested << endl;
for( int i=buffers.size(); i<Nrequested; i++){
if( buffer_pool.empty() ) break;
buffers.push_back( buffer_pool.back() );
buffer_pool.pop_back();
}
}
//-------------------------------------------------------------
// ReturnBuffers
//-------------------------------------------------------------
void hdRDMA::ReturnBuffers( std::vector<hdRDMAThread::bufferinfo> &buffers )
{
std::lock_guard<std::mutex> grd( buffer_pool_mutex );
for( auto b : buffers ) buffer_pool.push_back( b );
}
//-------------------------------------------------------------
// SendFile
//-------------------------------------------------------------
void hdRDMA::SendFile(std::string srcfilename, std::string dstfilename, bool delete_after_send, bool calculate_checksum, bool makeparentdirs)
{
// This just calls the SendFile method of the client hdRDMAThread
if( hdthr_client == nullptr ){
cerr << "ERROR: hdRDMA::SendFile called before hdthr_client instantiated." << endl;
return;
}
hdthr_client->SendFile( srcfilename, dstfilename, delete_after_send, calculate_checksum, makeparentdirs);
}
//-------------------------------------------------------------
// Poll
//
// Check for closed connections and release their resources.
// This is called periodically from main().
//-------------------------------------------------------------
void hdRDMA::Poll(void)
{
auto t_now = high_resolution_clock::now();
duration<double> delta_t = duration_cast<duration<double>>(t_now - t_last);
double t_diff = delta_t.count();
if( t_diff >=10.0 ){
//auto Ndiff = Ntransferred - Ntransferred_last;
//double rate_GB_per_sec = (double)Ndiff/t_diff/1.0E9;
//cout << "=== " << rate_GB_per_sec << " GB/s -- received " << Ndiff/1000000000 << "GB in last " << t_diff << "sec" << endl;
t_last = t_now;
Ntransferred_last = Ntransferred;
}
// Look for stopped threads and free their resources
std::lock_guard<std::mutex> lck( threads_mtx );
for( auto t : threads ){
if( t.second->stopped ){
t.first->join();
delete t.second;
threads.erase( t.first );
}
}
}