市面数据库调研（二）

承接上文对剩余5类数据库进行Overview（资料来源于某次调研英文展示，此部分尚未翻译并展开叙述，之后会补以新博文）

Key-value database

What :
a data storage paradigm designed for storing, retrieving, and managing associative arrays, a data structure more commonly known today as a dictionary or hash table.

Concept :
schema-free
Sharding;Distributed;CAP;BASE

Why :

• Reduce application server CPU and memory pressure
• Reduce IO read operations and IO stress
• The extensibility of relational database is not strong and it is difficult to change the table structure

Feature :

• Flexible data modeling
• Fast write performance
• Fast query performance

Use for :

• Session
• Distributed cache(High frequency data)(CAP)(BASE)
• A distributed lock

Comparison :

Storage : map,tree,…

Implement :
DynamoDB;Riak KV;Oracle NoSQL Database(table-style);
Oracle Berkeley DB(ACID);ArangoDB;FoundationDB
Memcached;Redis;Hazelcast(ACID)
LevelDB;RocksDB

Problem :

• support OLTP poorly

Document-oriented database

What :
a computer program designed for storing, retrieving and managing document-oriented information, also known as semi-structured data.

Why(Comparison with Key-value) :

• Data transparency(extract metadata that the database engine uses for further optimization)
• Designed to offer a richer experience with modern programming techniques

Concept :
Collections;Document
Inverted index

Use for :

• Search engines
• Log

Comparison :

Encoding type : XML;JSON;YAML;BSON;PDF;WORD;EXCEL

Implement :
MongoDB;DynamoDB;CouchDB
Elasticsearch
Terrastore;RavenDB(ACID);Thrudb
OrientDB

Column-oriented database

What :
a database management system (DBMS) that stores data tables by column rather than by row.

Feature :

• Extremely high loading speed

• For large data sets rather than small data sets

• Efficient storage space utilization

• High compression rate

• High CPU and memory utilization

• Don’t need the view

• Invisible index

• Efficient projection

• Efficient relational computing

• Efficient aggregation operations

• Efficient use of cache

Use for :
Data warehouse(Multiple data sources，Multidimensional data statistics，Large amount of stored data，More queries and fewer updates)

Implement :
HBase;Cassandra;Hypertable

Problem :

• Not suitable for scanning small amounts of data
• Not suitable for real-time operations with deletes and updates

Time-series database

What :
A time series database (TSDB) is a software system that is optimized for handling time series data, arrays of numbers indexed by time (a datetime or a datetime range).

if we don’t use TSDB,maybe hitory table is suitable.

Why :

• Large data size(important data history)(CRud)
• Availability(Time operation)
• Relational database primary key and foreign key support poorly

Use for :

• Monitor
• Real-time data processing(IOT)

Feature :

• Write more and read less
• Sequential insert
• Update less
• Bulk delete

---

• Distributed
• Time index
• Time function
• Time accuracy
• Timeliness
• Data compression

Storage engine :
HDFS, Hbase,Cassandra,LevelDB,…

Implement :

• InfluxDB,OpenTSDB,KDB+,Prometheus,KairosDB,HiTSDB
• Druid,Pinot
• RRDtool, Graphite

Focus :

• Scalability, Query language, Downsampling, High compression ratio
• OLAP
• Visualization

Graph database

What :
a database that uses graph structures for semantic queries with nodes, edges and properties to represent and store data.

Feature :

• Better at handling relationships than relational databases
• Graph theory algorithm (shortest path calculation, concentration calculation,…)

Use for :

• OLTP
• OLAP

---

• Search engine
• Social network
• Financial fraud prevention
• Risk early warning
• Business relationship
• Knowledge graph (emerging, high business abstraction, few landing except social network and search engine, more attempts in financial field)

Storage :

• Native graph (adjacency list, adjacency matrix) (good at traversal)
• Non-native graph (MySQL,…)(good at non-traversal operation)

Processing engine :
Cassovary,Pegasus,Giraph

Comparison :
Implement :
Neo4j（Neo4j Bloom）,AllegroGraph
FlockDB，GraphDB

Problem :

• not applicable to large-scale data
• does not apply to binary stored data
• not suitable for large amounts of event-driven data
• super node problem
• distributed large graph problem