Forecasting backup storage consumption

Potreba po rezervnem kopiranju oziroma arhiviranju podatkov v svetu narašča. Podjetja potrebujejo za normalno delovanje hraniti vedno več informacij. Hranjenje teh podatkov lahko predstavlja velik strošek, zato hočemo pomnilniško kapaciteto držati na ravni, ki zadovolji naše potrebe in hkrati ni predimenzionirana. S pomočjo podatkovnega rudarjenja želimo napovedati trende porabe pomnilniških kapacitet. Najprej smo pridobili podatke iz dveh različnih okolij za arhiviranje in jih shranili v podatkovno bazo. To nam je omogočilo hitro združevanje in upravljanje s podatki. Podatke smo analizirali z metodami linearne regresije, linearne regresije po kosih in k-najbližjih sosedov. Za najbolj zanesljivo metodo za napovedovanje trendov se je izkazala linearna regresija po kosih. Čeprav so rezultati dovolj dobri za uvedbo metode v produkcijo, moramo biti previdni, saj sta se analizirani okolji izkazali za zelo različni, kar neposredno vpliva na zanesljivost napovedi.Storage needs for archiving data are increasing. Companies need to store more and more data to function normally. Storing this data can be costly, that is why we want to provide sufficient storage capacity to meet the demands and not exceed them which brings additional costs. With the help of data mining we are trying to forecast trends in storage consumption. We acquired data from two environments for archiving and saved them to a database. We analysed data consumption trends with linear regression, piecewise linear regression and k-nearest neighbours. Piecewise linear regression proved to be the most accurate and reliable. Even though results are good enough to be implemented into production, we should be cautious as the two environments have different characteristics and this influences the forecasting