Getting started with Azure storage: Blobs vs Tables vs SQL Azure

Question

It's quite a topic, blobs vs tables vs SQL, and despite all I read so far I still can't find some proper reasoning on what to use when.

We have a multi-tenant SaaS web-application which we are about to move to Azure. We use an SQL Server 2008 database. We store documents and log information that belongs to the documents. Kinda like dropbox does.

The forums state that you better use Azure Tables when you are considering "large" objects. We typically store hundreds of documents per user where the size of the documents vary from 5kb to 30mb where the vast majority will be around 1MB?

Are there some ground rules when to go for Blobs, Tables, Sql? I already learned that I shouldn't store my documents in SQL since it is too expensive. But when does it get "beneficial" to store the documents in Blobs and when would I be better of with tables? Is there some kind of formula like :

if (objects * MB/object * objectrequested > y) then blobs, else tables

Answer 1

I think Igorek has addressed your SQL Azure concerns. You seem to still have questions about Tables vs Blobs, though.

In your case using Table storage would be annoying. Each property/column in ATS can be at most 64KB, so you would have to split the documents across multiple properties and then reassemble them. There is also a limit of 4MB per entity, which would be a problem. Blob storage has neither of these limitations.

I would tend to use Azure Table Storage when you have smallish entities with many properties that need to be stored and queried separately. So it works wells for stored objects, or small documents with lots of metadata.

Blob storage works better for things without a ton of metadata. It's good for things that might work well as files on a filesystem.

Answer 2

I would store documents themselves in the Azure Blob storage (not table storage). Outside of the fact that it is pretty expensive to store documents in a SQL Azure database that charges a penny per meg (or less depending on volume), SQL database is generally not a good place for documents. SQL is a relational database that provides benefits of ability to do queries, joins, etc. There is usually no benefit to storing large documents or images in a SQL database, especially when there is a highly scalable central storage system that is pretty cheap to store/access.

Now, if you need to search thru the documents themselves, I'd use something like Lucene.NET to provide a search capability for document-based repository.

HTH