Why is copying a stream and then deserializing using a BinaryFormatter faster than just deserializing

Question

This code takes about 8 seconds with a stream containing about 65K coming from a blob in a database

private string[] GetArray(Stream stream)
{
    BinaryFormatter binaryFormatter = new BinaryFormatter();
    object result = binaryFormatter.Deserialize(stream);
    return (string[])result;
}

This code takes a few milliseconds:

private string[] GetArray(Stream stream)
{
    BinaryFormatter binaryFormatter = new BinaryFormatter();
    MemoryStream memoryStream = new MemoryStream();
    Copy(stream, memoryStream);
    memoryStream.Position = 0;
    object result = binaryFormatter.Deserialize(memoryStream);
    return (string[])result;
}

Why?

Answer 1

So you say the problem disappears when the database is taken out of the equation. Here is my theory:

BinaryFormatter reads from the stream in tiny increments. It has to read as little as possible so that it does not accidentally swallow a few bytes after the serialized object. That means it is issuing tons of read commands (I verified this with Reflector).

Probably, every read of the blob stream is causing a network roundtrip (or some other major overhead). That gives you millions of roundtrips if using BinaryFormatter right away.

Buffering first causes the network to be utilized more efficiently because the read buffer size is much bigger.