INTERNATIONAL DATABASE ENGINEERING AND APPLICATIONS SYMPOSIUM A Data Masking Technique for Data Warehouses Ricardo Jorge Santos & Marco Vieira CISUC – DEI – FCTUC University of Coimbra - Portugal Jorge Bernardino CISUC – DEIS – ISEC Polytechnic Intitute of Coimbra - Portugal ISEL, Lisbon – September/2011 Agenda Background Motivation MOBAT Optimizing Features Experimental Results Conclusions & Future Work Agenda Background Motivation MOBAT: A MOD Based Data Masking Technique Optimization Features Experimental Results Conclusions and Future Work Ricardo J. Santos – A Data Masking Technique for Data Warehouses – IDEAS 2011 – ISEL, Lisbon – September/2011 2 Agenda Background Motivation MOBAT Optimizing Features Experimental Results Conclusions & Future Work Security Concerns in Data Warehousing A Data Warehouse (DW) is a critical asset for many enterprises Stores all relevant historical and current business information needed for supporting decision making (sensitive data) Main targets for stealing or compromising sensitive data Attack rate and complexity has increased in the recent past Ricardo J. Santos – A Data Masking Technique for Data Warehouses – IDEAS 2011 – ISEL, Lisbon – September/2011 3 Agenda Background Motivation MOBAT Optimizing Features Experimental Results Conclusions & Future Work Data Security Domains Data Confidentiality: Only the right users should access the right data Data Integrity: Data should always be correct, authentic and consistent Data Availability: User should always be able to access data whenever needed Ricardo J. Santos – A Data Masking Technique for Data Warehouses – IDEAS 2011 – ISEL, Lisbon – September/2011 4 Agenda Background Motivation MOBAT Optimizing Features Experimental Results Conclusions & Future Work Data Privacy Issues in Today’s DWs (Our Focus) Masking solutions are not considered an acceptable solution Encryption techniques introduce too much overheads Storage Space Data Loading Time Query Response Time Ricardo J. Santos – A Data Masking Technique for Data Warehouses – IDEAS 2011 – ISEL, Lisbon – September/2011 5 Agenda Background Motivation MOBAT Optimizing Features Experimental Results Conclusions & Future Work Data Privacy Issues in Today’s DWs (Our Focus) Important feature: Facts in DW’s are mainly numerical-based columns! Ricardo J. Santos – A Data Masking Technique for Data Warehouses – IDEAS 2011 – ISEL, Lisbon – September/2011 6 Agenda Background Motivation MOBAT Optimizing Features Experimental Results Conclusions & Future Work MOBAT – MOd BAsed data masking Technique for DWs MOBAT System Architecture Ricardo J. Santos – A Data Masking Technique for Data Warehouses – IDEAS 2011 – ISEL, Lisbon – September/2011 7 Agenda Background Motivation MOBAT Optimizing Features Experimental Results Conclusions & Future Work MOBAT – MOd BAsed data masking Technique for DWs Suppose table T => set of N numerical columns Ci = {C1, C2, C3, …, CN) to mask; total set of M rows Rj = {R1, R2, R3, …, RM). Each value to mask in the table identified as a pair (Rj, Ci) Rj and Ci respectively represent the row and column to which the value refers Each new masked value (Rj, Ci)’ is obtained by applying the following formula (1) for row j and column i of table T: (Rj, Ci)’ = (Rj, Ci) – ((K3, j MOD K1) MOD K2, i) + K2, i The inverse formula (2) for retrieving the original value is: (Rj, Ci) = (Rj, Ci)’ + ((K3, j MOD K1) MOD K2, i) – K2, i Ricardo J. Santos – A Data Masking Technique for Data Warehouses – IDEAS 2011 – ISEL, Lisbon – September/2011 8 Agenda Background Motivation MOBAT Optimizing Features Experimental Results Conclusions & Future Work MOBAT – Example Dataset Supposing K1 = 7432, K2,1 = 34 and K2,2 = 17252 Ricardo J. Santos – A Data Masking Technique for Data Warehouses – IDEAS 2011 – ISEL, Lisbon – September/2011 9 Agenda Background Motivation MOBAT Optimizing Features Experimental Results Conclusions & Future Work MOBAT – Example Dataset Supposing K1 = 9264, K2,1 = 12 and K2,2 = 78254 Ricardo J. Santos – A Data Masking Technique for Data Warehouses – IDEAS 2011 – ISEL, Lisbon – September/2011 10 Agenda Background Motivation MOBAT Optimizing Features Experimental Results Conclusions & Future Work MOBAT – Querying Using TPC-H benchmark with four numerical fact columns (i = 4) (L_Quantity, L_ExtendedPrice, L_Tax and L_Discount) masked by MOBAT New column L_KeyK3 for the j rows of the LineItem table, as the K3, j key K1=9342 K2, L_Quantity=12 K2, L_ExtendedPrice=51234 K2, L_Tax=6 SELECT SUM(L_ExtendedPrice * L_Discount) AS Total_Revenue K2, L_Discount=4 FROM LineItem WHERE L_ShipDate>=TO_DATE('1994-01-01','YYYY-MM-DD') AND L_ShipDate<TO_DATE('1995-01-01','YYYY-MM-DD') AND L_Discount BETWEEN 0.05 AND 0.07 AND L_Quantity<24 SELECT SUM((L_ExtendedPrice+MOD(MOD(L_KeyK3,9342),51234)-51234) * (L_Discount+MOD(MOD(L_KeyK3,9342),4)-4)) AS Total_Revenue FROM LineItem WHERE L_ShipDate>=TO_DATE('1994-01-01','YYYY-MM-DD') AND L_ShipDate<TO_DATE('1995-01-01','YYYY-MM-DD') AND (L_Discount+MOD(MOD(L_KeyK3,9342),4)-4) BETWEEN 0.05 AND 0.07 AND (L_Quantity+MOD(MOD(L_KeyK3,9342),12)-12)<24 Ricardo J. Santos – A Data Masking Technique for Data Warehouses – IDEAS 2011 – ISEL, Lisbon – September/2011 11 Agenda Background Motivation MOBAT Optimizing Features Experimental Results Conclusions & Future Work MOBAT – Optimizing Features & Performance The inclusion of K3,j requires additional storage space K3,j can be created in several ways, all with different impact in performance: Simply adding a new column to the previous existing fact table Recreating the fact table including K3,j from the start Using a 128-bit integer column already existing in the fact table (typically can be the primary key column) Ricardo J. Santos – A Data Masking Technique for Data Warehouses – IDEAS 2011 – ISEL, Lisbon – September/2011 12 Agenda Background Motivation MOBAT Optimizing Features Experimental Results Conclusions & Future Work Experimental Evaluation 2.8GHz CPU, 2GB RAM (512MB for Oracle SGA), 1.5TB SATA HD Oracle 11g DBMS One standard benchmark and one real-world DW TPC-H Decision Support Benchmark with 1GB and 10GB scale Real-world Sales DW (2GB storage size) Ricardo J. Santos – A Data Masking Technique for Data Warehouses – IDEAS 2011 – ISEL, Lisbon – September/2011 13 Agenda Background Motivation MOBAT Optimizing Features Experimental Results Conclusions & Future Work Experimental Evaluation Ricardo J. Santos – A Data Masking Technique for Data Warehouses – IDEAS 2011 – ISEL, Lisbon – September/2011 14 Agenda Background Motivation MOBAT Optimizing Features Experimental Results Conclusions & Future Work Experimental Evaluation Ricardo J. Santos – A Data Masking Technique for Data Warehouses – IDEAS 2011 – ISEL, Lisbon – September/2011 15 Agenda Background Motivation MOBAT Optimizing Features Experimental Results Conclusions & Future Work Experimental Evaluation Ricardo J. Santos – A Data Masking Technique for Data Warehouses – IDEAS 2011 – ISEL, Lisbon – September/2011 16 Agenda Background Motivation MOBAT Optimizing Features Experimental Results Conclusions & Future Work Conclusions Our technique decreases data storage space and processing overheads, while still proving a significant level of security Transparent method with minimal network bandwidth consumption overheads, due to only rewriting queries Extremely easy and simple to implement in any DBMS / DW, with low costs Querying the database directly will produce only realistic results (stored data is masked at all times) Ricardo J. Santos – A Data Masking Technique for Data Warehouses – IDEAS 2011 – ISEL, Lisbon – September/2011 17 Agenda Background Motivation MOBAT Optimizing Features Experimental Results Conclusions & Future Work Future Work Developing the technique for also masking alphanumeric values Assess its security strength in comparison with other solutions Developing the technique for increasing its security strength Using higher-sized keys Enabling data integrity checks Implementing false data injection Ricardo J. Santos – A Data Masking Technique for Data Warehouses – IDEAS 2011 – ISEL, Lisbon – September/2011 18 INTERNATIONAL DATABASE ENGINEERING AND APPLICATIONS SYMPOSIUM A Data Masking Technique for Data Warehouses THANK YOU! Questions and Comments? Ricardo Jorge Santos lionsoftware.ricardo@gmail.com ISEL, Lisbon – September/2011 19