Windows Azure - Automatic Load Balancing - partitioning

Question

I was going through some videos. I found that Windows Azure will group the blobs into partitions based on the partition key and will Automatically Load Balance these partitions on their servers. The partition key for a blob is blob name. Using the blob name, azure will automatically do partitions.

Now, My question is that Can I able to make the azure to do partitions based on the Container Name. I wanted my partition key to be container name.

For example, I have a storage account. In that I have 2 containers named container1 and container2. In container1, I have 1000 files named 1.txt, 2.txt, 3.txt, ......., 501.txt, 502.txt, ..... 999.txt, 1000.txt and in container2, I have another 1000 files named 1001.txt, 1002.txt, 1003.txt, ......., 1501.txt, 1502.txt, ..... 1999.txt, 2000.txt

Now, Will Windows Azure will generate 2000 partitions based on the blob name and serve me through several servers???

Won't it be better if Azure partitions based on the Container name? container1 on one server and conatiner2 on another.

Answer 1

That depends on what you mean by better. If you are paying for the individual servers and hard drives, I might agree with you - but with azure storage you just pay for the size of the space you use. If they choose to spread it across multiple machines to get the best perf (target blob access perf is "Up to 60 MBytes/sec") then it doesn't matter with your bill.

Just curious, why do you think having more partions is bad? To me it seems like with file access it is a good thing.

If I have 2000 files that I want to provide the most scalable access to, I would think having those 2000 files spread across 2000 machines/hard drives would be the best performing and scalable solution ... not 2 servers/hard drives (and what happens if that doubles to 4000?).

I found this blob article pretty informative (especially the area under the Partitions and Scalability and Performance Targets sections toward the bottom):Windows Azure Storage Abstractions and their Scalability Targets