Icepack registry "TimeOut" exception with heavy load

[eaglecn]

We've got two servers, each deploys a icepacknode which manages it's local service. Only one icepackregistry is used. We increase concurrent client number gradually and when exceeds some number, checkedCast throw Timeout Exception
==============
statement :
query =
IcePack::QueryPrx::checkedCast(ic->
stringToProxy
("IcePack/Query"));

Exception:
ConnectionI.cpp:127: Ice::ConnectTimeoutException:
timeout while establishing a connection
==============
original code is following:

        for (i = 0; i < 2; i++) {
            Ice::PropertiesPtr props1 = Ice::createProperties();
            props1->setProperty("Ice.Default.Locator",
                                "IcePack/Locator:default -p 12000  -t 100 -h "
                                + reg[i]);
            props1->setProperty("IcePack.Registry.Client.Endpoints",
                                "default -p 12000 -t 100 -h " + reg[i]);
            props1->setProperty("Ice.MessageSizeMax", "102400");
 
            props1->setProperty("Ice.Warn.Leaks", "0");
            try
            {
             ic = Ice::initializeWithProperties(argc, argv, props1);
            }
            catch(const Ice::ConnectTimeoutException &ex){
              errorStr = ex.ice_name();
              continue;
            }
            try {
                query =
                    IcePack::QueryPrx::checkedCast(ic->
                                                   stringToProxy
                                                   ("IcePack/Query"));
                int j = 0;
                registerflag = true;
                for (j = 0; j < 20; j++) {
                    try {
                        base =
                            query->
                            findObjectByType("::NetDB::MBatchStatement");
                        batch = MBatchStatementPrx::checkedCast(base);
                        if (batch) {
                            flag = true;
                            break;
                        }
                    }
                    catch(const Ice::Exception & ex) {
                        cout <<"service========"<< ex << endl;
                    }
                }
               if (j != 20)
                    break;
            }
            catch(const Ice::Exception & ex) {
                errorStr = ex.ice_name();
                cerr<<"registry=========="<<ex<<endl;
                ic->destroy();
            }
        }

when a server is down, the error will appear more quickly.

[matthew]

You have the timeout set to 100ms, which isn't a very long time -- especially once the server gets loaded... What do you find suprising in this scenario?