Access the file system efficiently

Access the file system efficiently (XAML)

The file system is a common source of performance issues. Accessing files can be expensive due to disk latency and memory/CPU cycles to store the data. When you want to access a large collection of files and you want to access property values other than the typical Name, FileType, and Path properties, access them by creating QueryOptions and calling SetPropertyPrefetch.

The SetPropertyPrefetch method can dramatically improve the performance of apps that display a collection of items obtained from the file system, such as a collection of images. The next set of examples shows a few ways to access multiple files.

The first example uses Windows.Storage.StorageFolder.GetFilesAsync to retrieve the name info for a set of files. This approach provides good performance, because the example accesses only the name property.

StorageFolder library = Windows.Storage.KnownFolders.PicturesLibrary;
IReadOnlyList<StorageFile> files = await library.GetFilesAsync(Windows.Storage.Search.CommonFileQuery.OrderByDate);
for (int i = 0; i < files.Count; i++)
    // do something with the name of each file
    string fileName = files[i].Name;

The second example uses Windows.Storage.StorageFolder.GetFilesAsync and then retrieves the image properties for each file. This approach provides poor performance.

StorageFolder library = Windows.Storage.KnownFolders.PicturesLibrary;
IReadOnlyList<StorageFile> files = await library.GetFilesAsync(Windows.Storage.Search.CommonFileQuery.OrderByDate);
for (int i = 0; i < files.Count; i++)
    ImageProperties imgProps = await files[i].Properties.GetImagePropertiesAsync();

    // do something with the date the image was taken
    DateTimeOffset date = imgProps.DateTaken;

The third example uses QueryOptions to get info about a set of files. This approach provides much better performance than the previous example.

// Set QueryOptions to prefetch our specific properties
var queryOptions = new Windows.Storage.Search.QueryOptions(CommonFileQuery.OrderByDate, null);
queryOptions.SetThumbnailPrefetch(ThumbnailMode.PicturesView, 100,
       new string[] {"System.Size"});

StorageFileQueryResult queryResults = KnownFolders.PicturesLibrary.CreateFileQueryWithOptions(queryOptions);
IReadOnlyList<StorageFile> files = await queryResults.GetFilesAsync();

foreach (var file in files)
    ImageProperties imageProperties = await file.Properties.GetImagePropertiesAsync();

    // Do something with the date the image was taken.
    DateTimeOffset dateTaken = imageProperties.DateTaken;

    // Performance gains increase with the number of properties that are accessed.
    IDictionary<String, object> propertyResults =
        await file.Properties.RetrievePropertiesAsync(
              new string[] {"System.Size" });

    // Get/Set extra properties here
    var systemSize = propertyResults["System.Size"];


If you're performing multiple operations on Windows.Storage objects such as Windows.Storage.ApplicationData.Current.LocalFolder, create a local variable to reference that storage source so that you don't recreate intermediate objects each time you access it.

Stream performance in C# and Visual Basic

Buffering between WinRT and .NET streams

There are many scenarios when you might want to convert a Windows Runtime stream (such as a Windows.Storage.Streams.IInputStream or IOutputStream) to a .NET stream (System.IO.Stream). For example, this is useful when you are writing a Windows Store app and want to use existing .NET code that works on streams with the Windows Runtime file system. In order to enable this, .NET APIs for Windows Store apps provides extension methods that allow you to convert between .NET and Windows Runtime stream types. For more info, see WindowsRuntimeStreamExtensions.

When you convert a Windows Runtime stream to a .NET stream, you effectively create an adapter for the underlying Windows Runtime stream. Under some circumstances, there is a runtime cost associated with invoking methods on Windows Runtime streams. This may affect the speed of your app, especially in scenarios where you perform many small, frequent read or write operations.

In order to speed up apps, the Windows Runtime stream adapters contain a data buffer. The following code sample demonstrates small consecutive reads using a Windows Runtime stream adapter with a default buffer size.

StorageFile file = await Windows.Storage.ApplicationData.Current
Windows.Storage.Streams.IInputStream windowsRuntimeStream = 
    await file.OpenReadAsync();

byte[] destinationArray = new byte[8];

// Create an adapter with the default buffer size.
using (var managedStream = windowsRuntimeStream.AsStreamForRead())

    // Read 8 bytes into destinationArray.
    // A larger block is actually read from the underlying 
    // windowsRuntimeStream and buffered within the adapter.
    await managedStream.ReadAsync(destinationArray, 0, 8);

    // Read 8 more bytes into destinationArray.
    // This call may complete much faster than the first call
    // because the data is buffered and no call to the 
    // underlying windowsRuntimeStream needs to be made.
    await managedStream.ReadAsync(destinationArray, 0, 8);

This default buffering behavior is desirable in most scenarios where you convert a Windows Runtime stream to a .NET stream. However, in some scenarios you may want to tweak the buffering behavior in order to increase performance.

Working with large data sets

When reading or writing larger sets of data you may be able to increase your read or write throughput by providing a large buffer size to the AsStreamForRead, AsStreamForWrite, and AsStream extension methods. This gives the stream adapter a larger internal buffer size. For instance, when passing a stream that comes from a large file to an XML parser, the parser can make many sequential small reads from the stream. A large buffer can reduce the number of calls to the underlying Windows Runtime stream and boost performance.

Note  You should be careful when setting a buffer size that is larger than approximately 80 KB, as this may cause fragmentation on the garbage collector heap (see Improve garbage collection performance). The following code example creates a managed stream adapter with an 81,920 byte buffer.

// Create a stream adapter with an 80 KB buffer.
Stream managedStream = nativeStream.AsStreamForRead(bufferSize: 81920);

The Stream.CopyTo and CopyToAsync methods also allocate a local buffer for copying between streams. As with the AsStreamForRead extension method, you may be able to get better performance for large stream copies by overriding the default buffer size. The following code example demonstrates changing the default buffer size of a CopyToAsync call.

MemoryStream destination = new MemoryStream();
// copies the buffer into memory using the default copy buffer
await managedStream.CopyToAsync(destination);

// Copy the buffer into memory using a 1 MB copy buffer.
await managedStream.CopyToAsync(destination, bufferSize: 1024 * 1024);

This example uses a buffer size of 1 MB, which is greater than the 80 KB previously recommended. Using such a large buffer can improve throughput of the copy operation for very large data sets (that is, several hundred megabytes). However, this buffer is allocated on the large object heap and could potentially degrade garbage collection performance. You should only use large buffer sizes if it will noticeably improve the performance of your app.

When you are working with a large number of streams simultaneously, you might want to reduce or eliminate the memory overhead of the buffer. You can specify a smaller buffer, or set the bufferSize parameter to 0 to turn off buffering entirely for that stream adapter. You can still achieve good throughput performance without buffering if you perform large reads and writes to the managed stream.

Performing latency-sensitive operations

You might also want to avoid buffering if you want low-latency reads and writes and do not want to read in large blocks out of the underlying Windows Runtime stream. For example, you might want low-latency reads and writes if you are using the stream for network communications.

In a chat app you might use a stream over a network interface to send messages back in forth. In this case you want to send messages as soon as they are ready and not wait for the buffer to fill up. If you set the buffer size to 0 when calling the AsStreamForRead, AsStreamForWrite, and AsStream extension methods, then the resulting adapter will not allocate a buffer, and all calls will manipulate the underlying Windows Runtime stream directly.



© 2017 Microsoft