Writing large number of records (bulk insert) to Access in .NET/C#

What is the best way to perform bulk inserts into an MS Access database from .NET? Using ADO.NET, it is taking way over an hour to write out a large dataset.

Note that my original post, before I "refactored" it, had both the question and answer in the question part. I took Igor Turman's suggestion and re-wrote it in two parts - the question above and followed by my answer.

Solution 1:

I found that using DAO in a specific manner is roughly 30 times faster than using ADO.NET. I am sharing the code and results in this answer. As background, in the below, the test is to write out 100 000 records of a table with 20 columns.

A summary of the technique and times - from best to worse:

1. 02.8 seconds: Use DAO, use DAO.Field's to refer to the table columns
2. 02.8 seconds: Write out to a text file, use Automation to import the text into Access
3. 11.0 seconds: Use DAO, use the column index to refer to the table columns.
4. 17.0 seconds: Use DAO, refer to the column by name
5. 79.0 seconds: Use ADO.NET, generate INSERT statements for each row
6. 86.0 seconds: Use ADO.NET, use DataTable to an DataAdapter for "batch" insert

As background, occasionally I need to perform analysis of reasonably large amounts of data, and I find that Access is the best platform. The analysis involves many queries, and often a lot of VBA code.

For various reasons, I wanted to use C# instead of VBA. The typical way is to use OleDB to connect to Access. I used an OleDbDataReader to grab millions of records, and it worked quite well. But when outputting results to a table, it took a long, long time. Over an hour.

First, let's discuss the two typical ways to write records to Access from C#. Both ways involve OleDB and ADO.NET. The first is to generate INSERT statements one at time, and execute them, taking 79 seconds for the 100 000 records. The code is:

public static double TestADONET_Insert_TransferToAccess()
{
  StringBuilder names = new StringBuilder();
  for (int k = 0; k < 20; k++)
  {
    string fieldName = "Field" + (k + 1).ToString();
    if (k > 0)
    {
      names.Append(",");
    }
    names.Append(fieldName);
  }

  DateTime start = DateTime.Now;
  using (OleDbConnection conn = new OleDbConnection(Properties.Settings.Default.AccessDB))
  {
    conn.Open();
    OleDbCommand cmd = new OleDbCommand();
    cmd.Connection = conn;

    cmd.CommandText = "DELETE FROM TEMP";
    int numRowsDeleted = cmd.ExecuteNonQuery();
    Console.WriteLine("Deleted {0} rows from TEMP", numRowsDeleted);

    for (int i = 0; i < 100000; i++)
    {
      StringBuilder insertSQL = new StringBuilder("INSERT INTO TEMP (")
        .Append(names)
        .Append(") VALUES (");

      for (int k = 0; k < 19; k++)
      {
        insertSQL.Append(i + k).Append(",");
      }
      insertSQL.Append(i + 19).Append(")");
      cmd.CommandText = insertSQL.ToString();
      cmd.ExecuteNonQuery();
    }
    cmd.Dispose();
  }
  double elapsedTimeInSeconds = DateTime.Now.Subtract(start).TotalSeconds;
  Console.WriteLine("Append took {0} seconds", elapsedTimeInSeconds);
  return elapsedTimeInSeconds;
}

Note that I found no method in Access that allows a bulk insert.

I had then thought that maybe using a data table with a data adapter would be prove useful. Especially since I thought that I could do batch inserts using the UpdateBatchSize property of a data adapter. However, apparently only SQL Server and Oracle support that, and Access does not. And it took the longest time of 86 seconds. The code I used was:

public static double TestADONET_DataTable_TransferToAccess()
{
  StringBuilder names = new StringBuilder();
  StringBuilder values = new StringBuilder();
  DataTable dt = new DataTable("TEMP");
  for (int k = 0; k < 20; k++)
  {
    string fieldName = "Field" + (k + 1).ToString();
    dt.Columns.Add(fieldName, typeof(int));
    if (k > 0)
    {
      names.Append(",");
      values.Append(",");
    }
    names.Append(fieldName);
    values.Append("@" + fieldName);
  }

  DateTime start = DateTime.Now;
  OleDbConnection conn = new OleDbConnection(Properties.Settings.Default.AccessDB);
  conn.Open();
  OleDbCommand cmd = new OleDbCommand();
  cmd.Connection = conn;

  cmd.CommandText = "DELETE FROM TEMP";
  int numRowsDeleted = cmd.ExecuteNonQuery();
  Console.WriteLine("Deleted {0} rows from TEMP", numRowsDeleted);

  OleDbDataAdapter da = new OleDbDataAdapter("SELECT * FROM TEMP", conn);

  da.InsertCommand = new OleDbCommand("INSERT INTO TEMP (" + names.ToString() + ") VALUES (" + values.ToString() + ")");
  for (int k = 0; k < 20; k++)
  {
    string fieldName = "Field" + (k + 1).ToString();
    da.InsertCommand.Parameters.Add("@" + fieldName, OleDbType.Integer, 4, fieldName);
  }
  da.InsertCommand.UpdatedRowSource = UpdateRowSource.None;
  da.InsertCommand.Connection = conn;
  //da.UpdateBatchSize = 0;

  for (int i = 0; i < 100000; i++)
  {
    DataRow dr = dt.NewRow();
    for (int k = 0; k < 20; k++)
    {
      dr["Field" + (k + 1).ToString()] = i + k;
    }
    dt.Rows.Add(dr);
  }
  da.Update(dt);
  conn.Close();

  double elapsedTimeInSeconds = DateTime.Now.Subtract(start).TotalSeconds;
  Console.WriteLine("Append took {0} seconds", elapsedTimeInSeconds);
  return elapsedTimeInSeconds;
}

Then I tried non-standard ways. First, I wrote out to a text file, and then used Automation to import that in. This was fast - 2.8 seconds - and tied for first place. But I consider this fragile for a number of reasons: Outputing date fields is tricky. I had to format them specially (someDate.ToString("yyyy-MM-dd HH:mm")), and then set up a special "import specification" that codes in this format. The import specification also had to have the "quote" delimiter set right. In the example below, with only integer fields, there was no need for an import specification.

Text files are also fragile for "internationalization" where there is a use of comma's for decimal separators, different date formats, possible the use of unicode.

Notice that the first record contains the field names so that the column order isn't dependent on the table, and that we used Automation to do the actual import of the text file.

public static double TestTextTransferToAccess()
{
  StringBuilder names = new StringBuilder();
  for (int k = 0; k < 20; k++)
  {
    string fieldName = "Field" + (k + 1).ToString();
    if (k > 0)
    {
      names.Append(",");
    }
    names.Append(fieldName);
  }

  DateTime start = DateTime.Now;
  StreamWriter sw = new StreamWriter(Properties.Settings.Default.TEMPPathLocation);

  sw.WriteLine(names);
  for (int i = 0; i < 100000; i++)
  {
    for (int k = 0; k < 19; k++)
    {
      sw.Write(i + k);
      sw.Write(",");
    }
    sw.WriteLine(i + 19);
  }
  sw.Close();

  ACCESS.Application accApplication = new ACCESS.Application();
  string databaseName = Properties.Settings.Default.AccessDB
    .Split(new char[] { ';' }).First(s => s.StartsWith("Data Source=")).Substring(12);

  accApplication.OpenCurrentDatabase(databaseName, false, "");
  accApplication.DoCmd.RunSQL("DELETE FROM TEMP");
  accApplication.DoCmd.TransferText(TransferType: ACCESS.AcTextTransferType.acImportDelim,
  TableName: "TEMP",
  FileName: Properties.Settings.Default.TEMPPathLocation,
  HasFieldNames: true);
  accApplication.CloseCurrentDatabase();
  accApplication.Quit();
  accApplication = null;

  double elapsedTimeInSeconds = DateTime.Now.Subtract(start).TotalSeconds;
  Console.WriteLine("Append took {0} seconds", elapsedTimeInSeconds);
  return elapsedTimeInSeconds;
}

Finally, I tried DAO. Lots of sites out there give huge warnings about using DAO. However, it turns out that it is simply the best way to interact between Access and .NET, especially when you need to write out large number of records. Also, it gives access to all the properties of a table. I read somewhere that it's easiest to program transactions using DAO instead of ADO.NET.

Notice that there are several lines of code that are commented. They will be explained soon.

public static double TestDAOTransferToAccess()
{

  string databaseName = Properties.Settings.Default.AccessDB
    .Split(new char[] { ';' }).First(s => s.StartsWith("Data Source=")).Substring(12);

  DateTime start = DateTime.Now;
  DAO.DBEngine dbEngine = new DAO.DBEngine();
  DAO.Database db = dbEngine.OpenDatabase(databaseName);

  db.Execute("DELETE FROM TEMP");

  DAO.Recordset rs = db.OpenRecordset("TEMP");

  DAO.Field[] myFields = new DAO.Field[20];
  for (int k = 0; k < 20; k++) myFields[k] = rs.Fields["Field" + (k + 1).ToString()];

  //dbEngine.BeginTrans();
  for (int i = 0; i < 100000; i++)
  {
    rs.AddNew();
    for (int k = 0; k < 20; k++)
    {
      //rs.Fields[k].Value = i + k;
      myFields[k].Value = i + k;
      //rs.Fields["Field" + (k + 1).ToString()].Value = i + k;
    }
    rs.Update();
    //if (0 == i % 5000)
    //{
      //dbEngine.CommitTrans();
      //dbEngine.BeginTrans();
    //}
  }
  //dbEngine.CommitTrans();
  rs.Close();
  db.Close();

  double elapsedTimeInSeconds = DateTime.Now.Subtract(start).TotalSeconds;
  Console.WriteLine("Append took {0} seconds", elapsedTimeInSeconds);
  return elapsedTimeInSeconds;
}

In this code, we created DAO.Field variables for each column (myFields[k]) and then used them. It took 2.8 seconds. Alternatively, one could directly access those fields as found in the commented line rs.Fields["Field" + (k + 1).ToString()].Value = i + k; which increased the time to 17 seconds. Wrapping the code in a transaction (see the commented lines) dropped that to 14 seconds. Using an integer index rs.Fields[k].Value = i + k; droppped that to 11 seconds. Using the DAO.Field (myFields[k]) and a transaction actually took longer, increasing the time to 3.1 seconds.

Lastly, for completeness, all of this code was in a simple static class, and the using statements are:

using System;
using System.Collections.Generic;
using System.Linq;
using System.Text;
using ACCESS = Microsoft.Office.Interop.Access; // USED ONLY FOR THE TEXT FILE METHOD
using DAO = Microsoft.Office.Interop.Access.Dao; // USED ONLY FOR THE DAO METHOD
using System.Data; // USED ONLY FOR THE ADO.NET/DataTable METHOD
using System.Data.OleDb; // USED FOR BOTH ADO.NET METHODS
using System.IO;  // USED ONLY FOR THE TEXT FILE METHOD

Solution 2:

Thanks Marc, in order to vote you I created an account on StackOverFlow...

Below is the reusable method [Tested on C# with 64 Bit - Win 7, Windows 2008 R2, Vista, XP platforms]

Performance Details: Exports 120,000 Rows in 4 seconds.

Copy the below code and pass the parameters... and see the performance.

• Just pass your datatable with the same schema, as of target Access Db Table.
• DBPath= Full path of access Db
• TableNm = Name of Target Access Db table.

The code:

public void BulkExportToAccess(DataTable dtOutData, String DBPath, String TableNm) 
{
    DAO.DBEngine dbEngine = new DAO.DBEngine();
    Boolean CheckFl = false;

    try
    {
        DAO.Database db = dbEngine.OpenDatabase(DBPath);
        DAO.Recordset AccesssRecordset = db.OpenRecordset(TableNm);
        DAO.Field[] AccesssFields = new DAO.Field[dtOutData.Columns.Count];

        //Loop on each row of dtOutData
        for (Int32 rowCounter = 0; rowCounter < dtOutData.Rows.Count; rowCounter++)
        {
            AccesssRecordset.AddNew();
            //Loop on column
            for (Int32 colCounter = 0; colCounter < dtOutData.Columns.Count; colCounter++)
            {
                // for the first time... setup the field name.
                if (!CheckFl)
                    AccesssFields[colCounter] = AccesssRecordset.Fields[dtOutData.Columns[colCounter].ColumnName];
                AccesssFields[colCounter].Value = dtOutData.Rows[rowCounter][colCounter];
            }

            AccesssRecordset.Update();
            CheckFl = true;
        }

        AccesssRecordset.Close();
        db.Close();
    }
    finally
    {
        System.Runtime.InteropServices.Marshal.ReleaseComObject(dbEngine);
        dbEngine = null;
    }
}

Solution 3:

You can use a KORM, object relation mapper that allows bulk operations over MsAccess.

database
  .Query<Movie>()
  .AsDbSet()
  .BulkInsert(_data);

or if you have source reader, you can directly use MsAccessBulkInsert class:

using (var bulkInsert = new MsAccessBulkInsert("connection string"))
{
   bulkInsert.Insert(sourceReader);
}

KORM is available from nuget Kros.KORM.MsAccess and it's opensource on GitHub

Solution 4:

Thanks Marc for the examples.
On my system the performance of DAO is not as good as suggested here:

TestADONET_Insert_TransferToAccess(): 68 seconds
TestDAOTransferToAccess(): 29 seconds

Since on my system the use of Office interop libraries is not an option I tried a new method involving the writing of a CSV file and then importing it via ADO:

    public static double TestADONET_Insert_FromCsv()
    {
        StringBuilder names = new StringBuilder();
        for (int k = 0; k < 20; k++)
        {
            string fieldName = "Field" + (k + 1).ToString();
            if (k > 0)
            {
                names.Append(",");
            }
            names.Append(fieldName);
        }

        DateTime start = DateTime.Now;
        StreamWriter sw = new StreamWriter("tmpdata.csv");

        sw.WriteLine(names);
        for (int i = 0; i < 100000; i++)
        {
            for (int k = 0; k < 19; k++)
            {
                sw.Write(i + k);
                sw.Write(",");
            }
            sw.WriteLine(i + 19);
        }
        sw.Close();

        using (OleDbConnection conn = new OleDbConnection(Properties.Settings.Default.AccessDB))
        {
            conn.Open();
            OleDbCommand cmd = new OleDbCommand();
            cmd.Connection = conn;

            cmd.CommandText = "DELETE FROM TEMP";
            int numRowsDeleted = cmd.ExecuteNonQuery();
            Console.WriteLine("Deleted {0} rows from TEMP", numRowsDeleted);

            StringBuilder insertSQL = new StringBuilder("INSERT INTO TEMP (")
                .Append(names)
                .Append(") SELECT ")
                .Append(names)
                .Append(@" FROM [Text;Database=.;HDR=yes].[tmpdata.csv]");
            cmd.CommandText = insertSQL.ToString();
            cmd.ExecuteNonQuery();

            cmd.Dispose();
        }

        double elapsedTimeInSeconds = DateTime.Now.Subtract(start).TotalSeconds;
        Console.WriteLine("Append took {0} seconds", elapsedTimeInSeconds);
        return elapsedTimeInSeconds;
    }

Performace analysis of TestADONET_Insert_FromCsv(): 1.9 seconds

Similar to Marc's example TestTextTransferToAccess(), this method is also fragile for a number of reasons regarding the use of CSV files.

Hope this helps.
Lorenzo