Heterogeneous high-energy-efficiency distributed database and application method

Abstract

The invention discloses a heterogeneous high-energy-efficiency distributed database and an application method. The database comprises a storage module, a read-write file system module and a database. A large number of IO operations are executed on the ARM chip, main CPU resources are not occupied, and the power consumption of the ARM chip is relatively reduced by one order of magnitude. Database reading and writing can be in accordance with hardware protocols, it is clearly known that hot data and temporary data need to be managed through different hardware in different scenes, a large amount of data does not need to be moved into a CPU main memory, and useless consumption is reduced. In addition, a file system is optimized, files are not read in a traditional mode, and performance loss caused by hardware characteristic dislocation is reduced. A large number of storage modules can be inserted into one service, but seamless expansion can be achieved in a virtual file system mode when the service is used. The database is modified at the bottom layer, compatibility is not damaged, direct migration of an existing system can be achieved, and program modification is not needed.

Description

technical field [0001] The invention relates to a heterogeneous high-energy-efficiency distributed database and an application method. Background technique [0002] Currently, there are two main types of mainstream business databases. One is mainly relational databases, which mainly realize fast data retrieval, data modification, and maintenance of transactional consistency. The other type of data is dominated by non-relational databases (NoSQL), and there are many types and scenarios of non-relational data. [0003] Most relational databases generally run under the x86 system and directly act on the general file system. Such databases generally do not have good support for distributed. At the same time, the architecture of this type of database is not sensitive to energy consumption. In the benchmark test state, the measured throughput is calculated at 500MBtyes / s, and the power consumption will be about 150w. [0004] From the analysis of the power consumption of the ru...

AI Technical Summary

Problems solved by technology

Compared with the demand for heavy IO from the database, it does not give full play to the advantages of a strong CPU. On the contrary, when waiting for IO for a long time, switching threads, processes and other operations will consume a lot of CPU power consumption and do a lot of useless work. At the operating system level Optimization is also extremely limited

[0005] Second, conventional databases do not optimize the file system used for storage

The second is that the database read-write module is based on the configuration of the current file system, and cannot make an effective single read volume for the hardware conditions.

It is very difficult to configure the block size of the read data volume, because due to the influence of the cache mechanism of the operating system, it is impossible to accurately utilize disk resources in one step

The third is that the database reads and sets the disk in the mode of passing through the read and write files, mainly synchronous reading and writing. When reading in parallel, it will be directly affected by the synchronous reading, and the same file cannot be retrieved at the same time, often the same table

Among them, when the data is retrieved twice, when the same large block of data is retrieved, it is impossible to merge the same retrieval and then split the data

[0006] In the third aspect, in the distributed mode of the database, the mutual data exchange needs to be synchronized through the network, but the network synchronization will occupy the bandwidth between the servers, and more distributed data needs to collect independent communication methods and CPU memory communication bus

Images