Recently, I found a great website that providing bunch of useful information about Windows Azure, especially in Benchmarking and Best Practice Guidance. Azurescope was developed by Research Engagements team in the eXtreme Computing Group of Microsoft Research as a resource for developers and application architects.

http://azurescope.cloudapp.net/

Best Practices

Azurescope discusses various best practice implementation to achieve best possible result. The best practices include:

• Deployment & Dynamic Scaling
• Design & Planning
• Windows Azure Storage
• SQL Azure
• Network Communication
• Testing & Development

Benchmark Test Cases

The benchmark contains 3 major categories test cases. Each test case describe clearly the scenario, key analysis, result that is showing on charts, how the benchmark works.

MicroBenchmark

Windows Azure Blob

• Blob Read Throughput
• Blob Write Throughput 
• Blob Read Latency 
• Blob Write Latency

Windows Azure Queue

• Queue Read Throughput 
• Queue Write Throughput 
• Queue Read Latency 
• Queue Write Latency

Windows Azure Table

• Table Read Throughput 
• Table Write Throughput 
• Table Read Latency 
• Table Write Latency

SQL Azure

• SQL Azure Read Throughput
• SQL Azure Write Throughput
• SQL Azure Read Latency
• SQL Azure Write Latency

Windows Azure Role

• Disk Read Throughput
• TDisk Write Throughput
• Memory Throughput
• CPU Throughput

Windows Azure Fabric

• Network Throughput
• Deployment Cost

Other Benchmarks

• Storage Type Comparison
• VM Size Comparison

End-to-end Application Benchmarks

1. Matrix Multiplication

These tests multiply two block square matrices of same dimensions to determine the suitability of Windows Azure platform for running embarrassingly parallel algorithms. The algorithms here partition the input matrix multiplication problem into smaller block multiplications. We employ b * b workers for running these tests where b is the number of blocks on one dimension of the input matrix.

2. Sorting

These tests sort few gigabytes of unsorted data stored in blob storage. The input unsorted data is split across multiple blobs where each blob contains multiple 100 byte records. Workers download these unsorted blobs, sort them locally and then upload sorted records back to the blob storage. The tests here try to determine whether Windows Azure is suitable for running parallel applications that require large data movements.

3. Sequence Alignment

This test implements a sequence matching algorithm so as to determine whether Windows Azure is suitable for running delightfully parallel algorithms.

Other Benchmarks

1. Storage Type Comparison

Charts on this page compare the maximum throughput achieved for the various storage types irrespective of the VM Size or values of other parameters on which the test was run. Please refer individual test cases for detailed information.

2. VM Size Comparison

Charts on this page compare the maximum throughput achieved for various storage types on various VM sizes irrespective of other test parameters. Please refer individual test cases for detailed information.

Code Samples

Apart from best practice and benchmark, AzureScope also provide a set of comprehensive code sample for each topic / component. Not only apply on .NET languages, the code sample also includes other languages such as PHP, Python, or Java.

General Windows Azure Code Sample

• Blob Upload and Download
• File Upload using BlobStream
• Parallel Blob Upload and Download
• CRUD Operation on Windows Azure Table
• Queue Message Add, Get and Delete
• Mounting and Unmounting Windows Azure Drive
• Use of retry policy provided by StorageClient to access Windows Azure storage
• Custom retry logic to access Windows Azure storage

Other Languages Code Samples

• Windows Azure IronPython Code Samples
• Windows Azure Java Code Samples
• Windows Azure PHP Code Samples

Hope this helps!