License: (CC 3.0) BY-NC-SA
intrudoction
current research
small file storage management
small file leading problem
-
document meta-data is stored in ram, couldn’t support processing of huge amount of small file fix: document archive & doecument merge
-
low io output fix: cache
-
low web bandwidth usage fix: request merge & request transmit delay
key tech for private cloud storage
layer and struct of private cloud storage
target of cloud storage
- large-scale document storage, and easy for liner expandtion
- realtime or nearly realtime access, bash support
- reliability, can handle single point or idc fault
- local file system like ui
- low cost, easy to maintain
four layers of cloud storage
- storage layer
- basic management layer
- application interface layer. include two type: web based REST protocal implementation, include web2.0 and dos like interface; specific open platform based api
- access layer
small cipher file storage management in private cloud
document-orented storage management basic principle for NoSQL feature
CAP theory
you can only satisfy 2 of 3 requirement in distribute computing for Consistency, Availability and tolerance of network Partition
BASE model
BASE model sacrifice consistency to get availability Basically Available, support partion failure Soft state, state can have asynchronization some time Eventually consistent, the final data should be consistent
prototype
dev platform
hadoop hdfs
it is suitable for large file, data stream, bash, not very satisfy small file storage
couchdb & mogondb
both are doc-oriented dbms, couchdb provide JSON data struct for REST interface to handle data process, mogondb store files as key-value pair.
RESTfull
REST (Representation State Transfer) is a group of constraint condition and principle, all program or design satisfy these condition and principle are RESTful. it use basic standard HTTP method.