Toshiba Digital Solutions explained the "GRIDDB" initiatives of the NOSQL type database management system (DBMS).Regarding GridDB, the company claims that overwhelming high performance can be achieved compared to other NOSQL databases that sell high speed.It is said to be adopted in systems that require high reliability and high performance, mainly in social infrastructure.
Big data / IoT specialized DB
GridDB was specialized for big data and IoT systems.Toshiba started development in 2011, commercialized as "GridStore" in 2013, and renamed GridDB for trademark issues.Since April 2021, it has also provided managed services in the cloud environment.
The background of Toshiba's development of a database (DB), Shinichiro Mochizuki, Senior Expert, New Business Division, ICT Solution Division, as follows.
"In order to maintain and manage the social infrastructure system handled by Toshiba, it was necessary to record and use the operating status of the system. In the past, other companies DB were used, but since the 2000s, system size.With the increase in the number of sensors and the increase in the number of sensors, the existing relational DB has become unable to meet the requirements. At that time, we considered NOSQL optimized for big data and IoT, but there are many open source software and the social infrastructure system.In order to use it, I was worried about reliability and developed DBMS myself. "
データベースとしての概要
The fact that the Toshiba Group had many engineers with DB development experience also supported the development of in -house.GridDB has four features.
"Time -series data adaptation" that efficiently and large -scale time -series data is processed in real time, "processing capacity" that is realized by incorporating a unique technology that handles petabite -scale data, in the event of a failure or when the server is expanded."Reliability and flexible expandability" that realizes non -stop operation, "agility and ease of use of development", which is realized not only by the NOSQL interface but also by SQL interface.
Mochizuki said, "GRIDDB has adopted a unique key container data model that has been extended for IoT data, and has taken good key values and good points.We have adopted an architecture that is ideal for data management. In addition, the event -driven engine has been used to eliminate excretion and other treatments and expected expectations. CPU can be fully rotated and the time -line data arrangement technology for maximally use memory.He uses "Time Series Data Placement Algorithm" (TDPA) to enable high -speed search. "
In TDPA, containers are created for each sensor, sensor data is generated in order of time, and stored in order of time.Even with multiple sensors, data at the same time is arranged close and stored.Also, if you search for the sensor and time on a condition, high -speed search is possible because the data to be searched is locally placed.It is said that the storage speed can be covered by making the most of the memory here.
GridDB can continue processing with another server backup data in the event of an automatic replication that holds data copies between servers.In addition, scale -out type DB has a weak point that the load is concentrated on the specific server if the balance of data is not placed, but in GridDB, Toshiba's unique autonomous data redistribution technology (ADDA) has a well -balanced and high -speed balance between the servers.Assuming that the data can be redoed.
自律データ再配置技術（ADDA）In ADDA, the master node collects node information and detects the imbalance of data between nodes and lack of backup.Apart from regular short -term synchronization, a long -term synchronization plan is determined from the current state (inbalancing), and the data redrow of long -term synchronization and short -term synchronization is executed.It is said that the background will be synchronized in the background while using a memory block and a DB update log in a range that does not give a load on the request process.After the data redoice is completed, rewrite the data placement information and switch access.
Mochizuki said, "The NOSQL interface that can be high -speed, high -throughput registration, search, and update, and the SQL interface supported by many technicians in the world will make it easier to develop and cooperate with applications."I say.You can use the NOSQL interface in mass data collection, and use the SQL interface in analysis and system cooperation.
This eliminates the need for shaping or integration of data by tool and batch processing between the DB for data accumulation and the DB for data analysis, and can perform real -time analysis."There are 380 kinds of DB worldwide, but it is rare to have NOSQL and SQL interfaces," says Mochizuki.
NoSQLとSQLの2つのインターフェースを持つIn the performance test, GridDB was better than other NOSQL type DB, which sells high performance, and has shown that it will be more performance as the number of nodes increases."GRIDDB is about 2.It demonstrated 5 times the performance, and when it was a lot of loading, it demonstrated 8 times the performance.Even if it is executed for a long time, the processing performance is maintained, and various workloads have performed 6 to 7 times the performance, "said Mochizuki."Compared to the relational DB, while maintaining the low load average and memory usage, the processing performance is about 13 times the processing performance by importing and extracting, and aggregation workloads, and the processing time is less than half."
