REST and ADO.NET interfaces
First need to create a storage account

• gives you a 256-bit secret key
• Will provide security for accessing the store

Abstractions

• Blobs (names files and metadata)
• Tables (structured set of entities)
• Queues
  

Concepts

• Storage account
• Container
• Blobs

Sharing policies is at the container level

Notes on Blob access:

• 50GB total blob storage limit for now.
• Put/Get/Delete API. 
• 64MB limit per blob in CTP with PutBlob.  larger must use PutBlock. 
• Block is up to 4MB.
• Can associate metadata with the blob: name value pairs, 8KB limit per blob
• Uploads can be resumed: blobs broken in to chunks.  PutBlock, PubBlockList API's. 
• Blocks not use will be GC'd. 
• GetBLockList can be retrieved for a blob. 
• MD5 can be put in the REST PutBlock and the storage system will use it.
• Not a real directory structure, but delimiters can be specified that simulate it.
• Update, Append and Copy coming in the future.

Notes on Tables:

• Massively scalable and durable (replicated at least 3 times)
• Can be accessed by ado.net 3.5 sp1
• Table is created in and scoped by account
• Table is a set of entities
• Entity is a set of properties
• Partition Key is exposed in the programming model, and helps to assist with scalability. 
  
• Partition gives entity locality (stored together - better query), and scalability (usage patterns monitored and used for load balancing)
• partition key should be used in query to optimize, otherwise all partitions must be examined.
• Entity can have up to 255 properties, including Partition Key, Row Key, and one system maintained version property
• No fixed schema
• Pertition Key and Row key string up to 64KB
• Table Entities access view ADO.NET DS, LINQ, and REST
• Insert/Update/Delete, query over tables
• Tables represented as class objects, adorn with DataServiceKey attribute
• Every account has a master table called "Tables"
• Use a DataServiceContext object to get entry point to the storage account
• LINQ works just like you expect
• Optimistic concurrency checking
• Not a relational database: no joins, foreign keys, etc
• For now, results are sorted by partitionkey and rowkey.  seconday indexes in the future
• Also, single entity transactions for now, future will be multiple entities

Notes on Queues:

• Provides reliable delivery of messages between workers and roles
• No limit on # of messages, but they can only live for a week
• Message <= 8KB.  Larger store a blob and reference it in a message
• Create/Clear/Delete, Inspect length of queue
• Messages: enqueue, dequeue, delete

Notes on durability, avail, scale

• Data replicated at least 3 times across different fault and upgrade domains in the same data center
• Geo-diversity will be after the CTP
• Efficient filover
• Automatic load balance hot data
• Monitor usage patterns fro optimization
• Caching of hot blobs, entities and queues