@@ -25,43 +25,44 @@ ParameterServerController::ParameterServerController(
2525 int numPorts = config.ports_num () + config.ports_num_for_sparse ();
2626
2727 if (config.nics ().empty ()) {
28- pservers_ .resize (numPorts);
28+ parameterServers_ .resize (numPorts);
2929 for (int i = 0 ; i < numPorts; ++i) {
3030 if (config.rdma_tcp () == " rdma" ) {
31- pservers_ [i].reset (
31+ parameterServers_ [i].reset (
3232 new ParameterServer2 (std::string (), config.port () + i, rdmaCpu++));
3333 rdmaCpu = rdmaCpu % onlineCpus;
3434 } else {
35- pservers_ [i].reset (
35+ parameterServers_ [i].reset (
3636 new ParameterServer2 (std::string (), config.port () + i));
3737 }
38- CHECK (pservers_[i]->init ()) << " Fail to initialize parameter server"
39- << config.port () + i;
38+ CHECK (parameterServers_[i]->init ()) << " Fail to initialize parameter "
39+ " server on port "
40+ << config.port () + i;
4041 }
4142 } else {
4243 str::split (config.nics (), ' ,' , &devices);
43- pservers_ .resize (devices.size () * numPorts);
44+ parameterServers_ .resize (devices.size () * numPorts);
4445 for (int i = 0 ; i < numPorts; ++i) {
4546 for (size_t j = 0 ; j < devices.size (); ++j) {
4647 if (config.rdma_tcp () == " rdma" ) {
47- pservers_ [i * devices.size () + j].reset (new ParameterServer2 (
48+ parameterServers_ [i * devices.size () + j].reset (new ParameterServer2 (
4849 getIpAddr (devices[j]), config.port () + i, rdmaCpu++));
4950 rdmaCpu = rdmaCpu % onlineCpus;
5051 } else {
51- pservers_ [i * devices.size () + j].reset (
52+ parameterServers_ [i * devices.size () + j].reset (
5253 new ParameterServer2 (getIpAddr (devices[j]), config.port () + i));
5354 }
54- CHECK (pservers_ [i * devices.size () + j]->init ())
55- << " Fail to initialize parameter server" << devices[j]
55+ CHECK (parameterServers_ [i * devices.size () + j]->init ())
56+ << " Fail to initialize parameter server with device " << devices[j]
5657 << config.port () + i;
5758 }
5859 }
5960 }
6061}
6162
62- ParameterServerController::~ParameterServerController () { this ->join (); }
63+ ParameterServerController::~ParameterServerController () { this ->wait (); }
6364
64- ParameterServerController* ParameterServerController::createByGflags () {
65+ ParameterServerController* ParameterServerController::createFromGflags () {
6566 ParameterServerConfig config;
6667
6768 config.set_nics (FLAGS_nics);
@@ -79,21 +80,21 @@ ParameterServerController* ParameterServerController::create(
7980}
8081
8182void ParameterServerController::start () {
82- LOG (INFO) << " pserver sizes : " << pservers_.size ();
83+ LOG (INFO) << " number of parameterServer instances: "
84+ << parameterServers_.size ();
8385 int i = 0 ;
84- for (const auto & pserver : pservers_ ) {
85- LOG (INFO) << " pserver started : " << i;
86- pserver ->start ();
86+ for (const auto & parameterServer : parameterServers_ ) {
87+ LOG (INFO) << " Starting parameterServer[ " << i << " ] " ;
88+ parameterServer ->start ();
8789 i++;
8890 }
8991}
9092
91- void ParameterServerController::join () {
92- LOG (INFO) << " pserver sizes : " << pservers_.size ();
93+ void ParameterServerController::wait () {
9394 int i = 0 ;
94- for (const auto & pserver : pservers_ ) {
95- LOG (INFO) << " pserver join : " << i;
96- pserver ->join ();
95+ for (const auto & parameterServer : parameterServers_ ) {
96+ LOG (INFO) << " Waiting parameterServer[ " << i << " ] " ;
97+ parameterServer ->join ();
9798 i++;
9899 }
99100}
0 commit comments