ANALYSIS OF RECORD VERSIONS RECONCILIATION PROCESSES IN NOSQL DATABASES

• Main Authors: Yu. A. Grigorev, А. D. Plutenko, А. V. Burdakov, E. V. Tsvyashchenko
• Format: Article
• Language: English
• Published: CRI «Electronics» 2017-11-01
• Series: Радиопромышленность
• Subjects: nosql; record versions; model of reconciliation; adequacy
• Online Access: https://www.radioprom.org/jour/article/view/265
• ISSN: 2413-9599; 2541-870X

At present, along with relational databases, the NoSQL databases have become widely used. They have high performance and  reliability,  because  the  data  are  stored  in  form  of  records  <key,  value>,  which  are  repeatedly  replicated.  However, absence of an explicit mechanism for conducting transactions and locks of records in these databases leads to a sharp increase in the number of versions of records in the database, while they are updated by several users. In this case, the user workload and time of reconciliation of these versions are growing up, the system performance decreases, and the  probability of version loss increases. The number of types of records stored in the database is large, and the number of versions of records cannot be measured at the stage of system operation. Therefore, the actual task is to estimate the number  of  versions  of  the  record  and  the  time  of  their  reconciliation  at  the  peak  load  on  the  system  at  the  stage  of  its design.  This  gives  a  possibility  to  make  a  right  decision  in  time.  In  this  case,  adequate  mathematical  models  are  an indispensable tool for analysis. The article discusses the process of processing (reconciliation) of the record versions. A simulation model has been developed for reconciling the versions of the record, which is updated in parallel by several users. The model allows estimating the time of agreement between the versions of the record and the number of versions simultaneously stored in the database. Based on the results of simulation, the system designer can give recommendations on  the  maximum  possible  number  of  users  (or  applications)  simultaneously  working  with  one  document  (a  database record). This is important if restrictions are placed on the time of document reconciliation. The process of preparation and carrying out of full-scale experiments in the cloud for analysis of model adequacy is described. The results of full-scale experiments show that the developed simulation model of the process of keeping the versions of the record is adequate. The average relative uncertainty in estimating the characteristics of the random number of versions of the record by the model is: for the mean value – 7,5% and the right confidence limit (RCI) –  3,0%. The average relative uncertainty of the estimation by the mean value of the update processing time is 0.22%, RCI – 5,6%.