将大型SQL Server CE数据库导出到XML文件的更快方法？

Question

What is a faster way to export large (25,000 rows) SQL Server CE database into XML file?

Here is what I currently use:

using (SqlCeConnection cn = new SqlCeConnection(strConnection))
{
    if (cn.State == ConnectionState.Closed)
        cn.Open();
    using (SqlCeCommand cmd = new SqlCeCommand(strCommand, cn))
    {
        SqlCeDataAdapter da = new SqlCeDataAdapter(cmd);
        DataSet ds = new DataSet();
        da.Fill(ds, "item");
        StreamWriter xmlDoc = new StreamWriter("Output.xml");
        ds.WriteXml(xmlDoc);
        xmlDoc.Close();
    }
}

It takes about 60 seconds inside emulator (Windows Mobile).

Also.. I am using Compact Framework 3.5 with SQL Server CE 3.5.

Current performance: 60 seconds entire code ~20 seconds for everything without ds.WriteXml(xmlDoc); , leaving ~40 seconds for ds.WriteXml(xmlDoc); .

Answer 1

Maybe something like this is more lightweight:

    private void ExportDataTable(SqlCeCommand cmd, SqlCeConnection conn)
    {
        cmd.Connection = conn;
        System.Data.DataTable table = new System.Data.DataTable();
        table.Locale = CultureInfo.InvariantCulture;
        table.Load(cmd.ExecuteReader());
        StreamWriter xmlDoc = new StreamWriter("Output.xml");
        table.WriteXml(xmlDoc);
    }

Answer 2

If the effort is worth it to you, the fastest method is probably to roll your own XML generation. The library implementation of XmlWriter() is loops within loops, with considerable generality.

Declare your own output stream specifying the buffer size to something reasonably large (but still sane). I have tested StringBuilder() before and append() to it millions of times with good performance. Other output options may not be as fast, but I would hope the StreamWriter gives good performance when buffersize if appropriate and it force you to build everything in memory.

Don't use da.Fill() -- replace with a SqlCeDataReader.

For each row. Generate the XML for a single row using logic that is coded as close as possible to the metal. Ie, Precalculate column index values, etc. instead of using column names within this loop, use hard-coded type conversions as needed. Don't loop through each column, put each column

Also, test by having the database generating XML output. Although I don't expect this to be the fastest, it is easy to try, and if it turns out to be fast you would never have discovered it without trying.