Essential Cloud Services

Blob Namespace

http://<account>.blob.core.windows.net/<container>/<blobname>
can be used as a hierarchical namespace
can be up to 50GB each
accessed via REST PUT/GET interface
includes continuation on upload scenario: blobs are broken up into blocks; if large file upload fails, it can resume without retrying successful blocks; PutBlockList commits the upload
under the covers, each blob consists of several immutable blocks (up to 4 Mb each)
if multiple blocks with the same ID are written, last in wins
blocks not associated with the blob on PutBlockList are garbage collected
future support: update, append, copy

Tables

provides structured scalable via ADO .NET data services
a table is a collectino of entities which have properties
every entity is uniquely identified by a partition key and the row key (composite identifier)
partition key: allows grouping of entities into table partition; exposed to allow control of the granularity of the table partitions
entities in the same partition will be stored together (entity locality); tables will be partitioned between servers based on cloud algorithm
choosing your partition key will be something of an art form; essentially, it needs to be a common subset of your query criteria but you want the load spread out across multiple partitions
basically, this is horizontal partitioning, with the implementation details abstracted away from the developer
an entity can have up to 255 properties, subject to the aforementioned mandatory properties
entities with a table can have heterogeneous properties
every account has a master table called “Tables”

Queues

provide reliable message delivery between web roles and worker roles
messages can by up to 8Kb in size
have an “invisibility time” on dequeue; this is the duration within which you must call delete for the message to be deleted
otherwise, the message becomes visible again, allowing the message processing to be attempted again

Storage Durability

all data is replicated at least 3 times: concept of fault and upgrade domains; the replicas are stored in the same datacenter but in different domains
failover and load balancing provided; hot blobs are also cached

A Technical Debtor