OTL library - SQL Server - C++ - Performance - visual-c++

Question

I have recently started testing the OTL library with the SQL Server using the Visual Studio 2013. My tests demonstrated that the performance of simple select statements against a 10000 count table is 40% slower than the performance of a similar .NET 4.0 test application. All tests were performed with all optimizations turned on for both platforms.
Both apps perform the following tasks:
Open the db connection
Create (and reserve space) for the container object.
Execute the select statement command.
For each record fetched from db
create an entity using the db(stream/reader) object
add the object to container
close
.NET C# app requires 0.5 secs to complete this task, while OTL-C++ app takes 0.7 secs to complete and I wonder if it is possible to optimize the C++ app to perform faster?
Snippet of C++ code:
#define OTL_ODBC_MSSQL_2008 // Compile OTL 4/ODBC, MS SQL 2008
#define OTL_CPP_11_ON
#define OTL_STL // Turn on STL features
#define OTL_ANSI_CPP // Turn on ANSI C++ typecasts
#define OTL_UNICODE // Enable Unicode OTL for ODBC
#include "otlv4.h"
class Employee
{
private:
int employeeId;
wstring regno;
wstring name;
wstring surname;
public:
Employee()
{
}
Employee(otl_stream& stream)
{
unsigned short _regno[32];
unsigned short _name[32];
unsigned short _surname[32];
if (!stream.is_null())
{
stream >> employeeId;
}
if (!stream.is_null())
{
stream >> (unsigned char*)_regno;
regno = (wchar_t*)_regno;
}
if (!stream.is_null()){
stream >> (unsigned char*)_name;
name = (wchar_t*)_name;
}
if (!stream.is_null()){
stream >> (unsigned char*)_surname;
surname = (wchar_t*)_surname;
}
}
int GetEmployeeId() const
{
return employeeId;
}
};
otl_connect db;
int main()
{
otl_connect::otl_initialize();
try
{
otl_connect::otl_initialize();
try
{
// connect
db.rlogon("DSN=SQLODBC");
// start timing
clock_t begin = clock();
otl_stream i(10000, "SELECT Id, Field1, Field2, Field3 FROM Test", db);
// create container
vector<Employee> employeeList;
employeeList.reserve(10000);
// iterate and fill container
while (!i.eof())
{
Employee employee(i);
employeeList.push_back(employee);
}
i.close();
// cleanup
size_t size = employeeList.size();
clock_t end = clock();
double elapsed_secs = double(end - begin) / CLOCKS_PER_SEC;
cout << "Total records:" << size << endl;
cout << "Time elapsed to read all records:" << elapsed_secs << endl;
}
catch (otl_exception& p){ // intercept OTL exceptions
cerr << p.msg << endl; // print out error message
cerr << p.stm_text << endl; // print out SQL that caused the error
cerr << p.sqlstate << endl; // print out SQLSTATE message
cerr << p.var_info << endl; // print out the variable that caused the error
}
db.logoff();
return EXIT_SUCCESS;
}

Answer 1

I don't think so, when you look the code source of otl, it does use ODBC api for SQL Server and is only optimized as a odbc top layer. The SQL Server .NET 4.0 will use the sql driver api instead of odbc api for performance reason.
Also if you don't preallocate your memory consumption, you will always loose to .NET and Java due to the SysAllocMem function call. It's like trying to measure 4000 call to SysAlloc vs 1 call to SysAlloc. Your performance issue is directly linked to those functions.