Cloud storage of virtual storage blog directory
Cloud computing storage
1. The conventional memory limit
(1) local store
Local storage
needs to increase by RAID redundancy to improve the reliability of the data
disk is not shared between the host, the host can not access across
IO bottlenecks and capacity, not smooth expansion
VM HA restricted
(2) external storage
External storage
need to purchase third-party storage equipment, the cost is very high
storage deployment of complex, difficult to maintain
compatibility problems
2. Distributed Storage
Traditional network storage system uses a centralized storage server to store all data, storage servers become the bottleneck of the system performance, reliability and security is the focus, and can not meet the needs of large-scale storage applications.
Distributed storage, is stores data on a plurality of separate devices.
Distributed network storage system uses a scalable system architecture using multiple storage servers share the load store, the server using the location information stored in the positioning, it will not only improve the reliability, availability and efficiency of the system to access, easy to expand further.
2.1 distributed storage characteristics
Distributed storage properties
Transparency: The procedure and the underlying physical storage locations transparent to the user
open: standard access interface operable, portable
scalability: based on distributed design, smooth expansion
availability: redundant data security
agreement : the communication failure preventing split brain cause the storage system
Distributed File System 2.2
Distributed Storage Distributed File System used (Distributed File System) network in the same physical storage resources of different computer management (local storage) through a network organized to form a single shared file system.
Distributed File System Based on C / S structure mode. When the client to access files without knowing their actual physical storage location, the distribution of files on multiple computers as using local storage as in front of the user.
On behalf of open source distributed file system
• Google File System
In order to store large amounts of data Google Search designed dedicated file system.
• Hadoop Distributed File System
Powered by Apache open source distributed file system, a high degree of fault tolerance to provide high throughput data access for applications on large data sets to support distributed computing.
• GlusterFS
Z RESEARCH mainly by the company responsible for the development, application in a cluster system, has good scalability. Modular design, easy to extend and configure.
3. Sangfor introduced virtual storage
Sangfor virtual storage (referred to aSAN) is based on GlusterFS distributed file system developed by the face of storage virtualization trend of a product.
Current aSAN integrated in the enterprise and desktop cloud cloud aCloud aDesk, within the network by integrating all the local hard disk cluster management server.
3.1 aSAN read cache, data tiering
ASAN performance using a variety of optimization techniques, including SSD write cache, SSD layering technology, the IO localization, link aggregation, etc., provide excellent performance as aSAN.
Read, write cache, data tiering using different implementation principle, its role isThe use of SSD caching disk read and write to accelerate and enhance the performance of virtual storage
3.2 Virtual Storage Networking
aSAN virtual storage is based on a distributed storage network, all hosts need to be connected via an IP network (Layer, multicast), set up to store communications network.
Basic requirements:
stable communication: the delay does not exceed 5ms, lossless
sufficient bandwidth: 10G LAN connection or Gigabit
Storing private network has three choices, namely polymerization no link, switch link aggregation single, double switch link aggregation.
(1) no link aggregation
No link aggregation: per host using a separate interconnect network port, the formation of private network is stored
Advantages: storing the traffic data and data isolation
disadvantages: storing private network without redundant, will lead to a link failure occurs corresponding to the host memory can not be used
(2) single-switch link aggregation
Switch Link Aggregation: each host using two separate storage network ports form a private network, the communication network is stored all access to the same port switcher (gigabit above), polymerization is carried out automatically between aDesk / aCloud host, without the need polymerization on the switch.
Advantages: double the bandwidth of link redundancy and has
drawbacks: the presence of single point of failure risk of the switch (switches do recommended stack)
(3) A double switch link aggregation
Dual Switch Link Aggregation: each host using two separate storage network ports form a private network, the communication network is stored respectively connected to two ports switcher (gigabit above), polymerization is carried out automatically between aDesk / aCloud host, without polymerization arranged on the switch.
Advantages: double the bandwidth, and redundant links and switch has
disadvantages: the need for increased network interface devices and switches