How to achive more 10 inserts per second with azure storage tables

Question

I write simple WorkerRole that add test data in to table. The code of inserts is like this.

var TableClient = this.StorageAccount.CreateCloudTableClient();
T         


        

Answer 1

To speed things up you should use batch transactions (Entity Group Transactions), allowing you to commit up to 100 items within a single request:

foreach (var item in myItemsToAdd)
{
    this.Context.AddObject(TableName, item);
}
this.Context.SaveChanges(SaveChangesOptions.Batch);

You can combine this with Partitioner.Create (+ AsParallel) to send multiple requests on different threads/cores per batch of 100 items to make things really fast.

But before doing all of this, read through the limitations of using batch transactions (100 items, 1 partition per transaction, ...).

Update:

Since you can't use transactions here are some other tips. Take a look at this MSDN thread about improving performance when using table storage. I wrote some code to show you the difference:

    private static void SequentialInserts(CloudTableClient client)
    {
        var context = client.GetDataServiceContext();
        Trace.WriteLine("Starting sequential inserts.");

        var stopwatch = new Stopwatch();
        stopwatch.Start();

        for (int i = 0; i < 1000; i++)
        {
            Trace.WriteLine(String.Format("Adding item {0}. Thread ID: {1}", i, Thread.CurrentThread.ManagedThreadId));
            context.AddObject(TABLENAME, new MyEntity()
            {
                Date = DateTime.UtcNow,
                PartitionKey = "Test",
                RowKey = Guid.NewGuid().ToString(),
                Text = String.Format("Item {0} - {1}", i, Guid.NewGuid().ToString())
            });
            context.SaveChanges();
        }

        stopwatch.Stop();
        Trace.WriteLine("Done in: " + stopwatch.Elapsed.ToString());
    }

So, the first time I run this I get the following output:

Starting sequential inserts.
Adding item 0. Thread ID: 10
Adding item 1. Thread ID: 10
..
Adding item 999. Thread ID: 10
Done in: 00:03:39.9675521

It takes more than 3 minutes to add 1000 items. Now, I changed the app.config based on the tips on the MSDN forum (maxconnection should be 12 * number of CPU cores):

  <system.net>
    <settings>
      <servicePointManager expect100Continue="false" useNagleAlgorithm="false"/>
    </settings>
    <connectionManagement>
      <add address = "*" maxconnection = "48" />
    </connectionManagement>
  </system.net>

And after running the application again I get this output:

Starting sequential inserts.
Adding item 0. Thread ID: 10
Adding item 1. Thread ID: 10
..
Adding item 999. Thread ID: 10
Done in: 00:00:18.9342480

From over 3 minutes to 18 seconds. What a difference! But we can do even better. Here is some code inserts all items using a Partitioner (inserts will happen in parallel):

    private static void ParallelInserts(CloudTableClient client)
    {            
        Trace.WriteLine("Starting parallel inserts.");

        var stopwatch = new Stopwatch();
        stopwatch.Start();

        var partitioner = Partitioner.Create(0, 1000, 10);
        var options = new ParallelOptions { MaxDegreeOfParallelism = 8 };

        Parallel.ForEach(partitioner, options, range =>
        {
            var context = client.GetDataServiceContext();
            for (int i = range.Item1; i < range.Item2; i++)
            {
                Trace.WriteLine(String.Format("Adding item {0}. Thread ID: {1}", i, Thread.CurrentThread.ManagedThreadId));
                context.AddObject(TABLENAME, new MyEntity()
                {
                    Date = DateTime.UtcNow,
                    PartitionKey = "Test",
                    RowKey = Guid.NewGuid().ToString(),
                    Text = String.Format("Item {0} - {1}", i, Guid.NewGuid().ToString())
                });
                context.SaveChanges();
            }
        });

        stopwatch.Stop();
        Trace.WriteLine("Done in: " + stopwatch.Elapsed.ToString());
    }

And the result:

Starting parallel inserts.
Adding item 0. Thread ID: 10
Adding item 10. Thread ID: 18
Adding item 999. Thread ID: 16
..
Done in: 00:00:04.6041978

Voila, from 3m39s we dropped to 18s and now we even dropped to 4s.