NOSQL

in Mid 80s, rise of relational : persistence , sql transactions, reporting

sql problem: assemble lots of table into one logic entity, impedance mismatch, 

in Mid 90s, rise of object database, argue endlessly

Integration database, relational dominance

web sites :

1 lots of traffic , scale things up, you bug bigger box

2 lots of small boxes, sql doesn't not work well around boxes.

so,

google --> bigtable

amazon --> dynamo

"NOSQL"

mongodb, couchdb,cassandra,hypetable,dynomite,hbase,

characteristics of nosql

non-relational, cluster-friendly, schema-less, open-source, 21st century web

data model vs relational model

Column-family	document	key-value	graph
cassandra	mongodb	redis	neo4j
hbase	couchdb	riak
	ravendb	project voldemort

 

 

 

1. graph
2. aggregate-oriented (domain-driven design: eric evans)

• key-value: hashmap persistent on disk : customer_id: 7231 metadata value= aggregate
• document : no schema , json like, transparent

　　anOrder["price"] * anOrder["quantity"] : implicit schema, assume "price" and "quantity", schema less, uhm, document=aggregate

 

• column-family

　　row-key +column family name+column key+column value

however, nothing is perfert.

rdbms == acid

nosql  == base 

graph : acid

you cann't afford to hold transaction that long

version stamp!!! offline lock: transaction boundary

consistency : 

• logical
• replication

CAP:

Partition + Consistency

Partition + Availability

Consistency <==spectrum==> Availability

Consistency <== ==> Response time

 

driver to nosql:

large scale data , easier development

Integation database === > application database (web service)

applications control access to data. ==>nosql, not integration into database!!!

 

for data warehouse

agile analytics 

 

Ployglot persistence: relational still play, choose appropriate db for your biz.

rapid response, data intensive