Database Reference
In-Depth Information
databases refer to general design concepts rather than strict categories, as features of
one type are often found in the other.
Key-Value Database
It goes without saying that Amazon.com attracts a large number of users, and, out
of necessity, the company has been on the cutting edge of application scalability. A
seminal paper about the nonrelational database technology behind Amazon's Dynamo
database, entitled “Amazon's Highly Available Key-Value Store,” describes a set of use
cases for which relational databases are not ideal. “For many services,” the paper states,
“such as those that provide best seller lists, shopping carts, customer preferences, session
management, sales rank, and product catalog, the common pattern of using a relational
database would lead to inefficiencies and limit scale and availability.”
2
When data vol-
umes grow, reading from and writing to a relational database while maintaining ACID
compliance can be computationally expensive.
A key-value store looks a lot like a big hash table. Each record in the database is an
object identified by a unique key. The value that is addressed by this key can basically
be anything: a string, a JSON or XML document, a binary blob, or a number of other
things. A key characteristic of these databases is that the system doesn't really know
anything about the data being stored; all access to data comes from the key. Key-value
stores are the right choice for applications that generally retrieve database information
based on a single key.
This design makes for very good performance for database writes. Another advan-
tage is that the design lends itself toward being easy to scale and replicate across a net-
work of machines. Because each record in the database is simply a value addressed by
a unique key, the data itself can reside on any machine on the network as long as the
system knows how it can be located. One way to scale this type of database is to keep
a secondary table of key-range to machine mappings. When a piece of data is defined,
this table may be referred to by the secondary table and the request sent to the proper
machine.
On the other hand, a general disadvantage of the design of a key-value store is that
data cannot be accessed by value. In other words, it is impossible to query a key-value
data store for all records that contain a particular set of values. The only way to query a
key-value database is by specifying a request by key or, in some cases, a range of keys.
Listing 3.2 demonstrates some examples of using a key-value store to query for data.
Listing 3.2
Organizing data by key using a key-value data store
> SET book:1 "Data Just Right"
> SET author:1 "Michael Manoochehri"
> GET book:1
"Data Just Right"
Search WWH ::
Custom Search