L-EncDB: A lightweight framework for privacy-preserving data queries in cloud computing

TL;DR: A new format-preserving encryption (FPE) scheme is constructed in this paper, which can be used to encrypt all types of character strings stored in database and is highly efficient and provably secure under existing security model.

Abstract: With the advent of cloud computing, individuals and organizations have become interested in moving their databases from local to remote cloud servers However, data owners and cloud service providers are not in the same trusted domain in practice For the protection of data privacy, sensitive data usually have to be encrypted before outsourcing, which makes effective database utilization a very challenging task To address this challenge, in this paper, we propose L-EncDB, a novel lightweight encryption mechanism for database, which (i) keeps the database structure and (ii) supports efficient SQL-based queries To achieve this goal, a new format-preserving encryption (FPE) scheme is constructed in this paper, which can be used to encrypt all types of character strings stored in database Extensive analysis demonstrates that the proposed L-EncDB scheme is highly efficient and provably secure under existing security model

Summary

1 Introduction

• The ever-increasing amount of valuable digital data both at home and in business needs to be protected, since its irrevocable loss is unacceptable.
• The advent of cloud storage motivates enterprises and organizations to outsource data storage to third-party cloud providers, as evidenced by many real-life case studies.
• In recent years, their popularity has increased dramatically.
• In the proposed L-EncDB system, the encryption and query are based on SQL.

1.1 Organization

• The rest of this paper proceeds as follows.
• In Section 3, the authors propose the system architecture and construction method for the L-EncDB system.
• Its security and performance analysis is also given in this section.

2.1 FPE

• The notion of FPE [1–4] has been proposed to generate ciphertext with the same format as plaintext while encrypting sensitive information.
• More specifically, FPE can keep data type and length in the ciphertext, therefore, without changing database structure and field type.
• There is no suitable character FPE solution to preserve both length and storage size of strings above.
• Such a character set is called “multi-byte character set”.
• Cycle-walking can not guarantee stable efficiency, which is impractical in most applications.

2.2 Privacy-preserving Database Encryption

• A number of research results [9–13] were proposed for privacy-preserving database encryption.
• To support query over encrypted numerical data, Hakan et al. [9] firstly presented a basic framework of how to ensure data security in “Database-As-Service” (DAS) model, in which a coarse query is executed by the database service provider.
• The proxy encrypts and decrypts all the data, and changes some query operators, while preserving the semantics of the query.
• In cloud computing, users are able to store, modify and retrieve data from anywhere in the world, as long as they have access to the Internet.
• CryptDB changes the database structure and stores the ciphertexts generating by different encryption methods.

2.3 Other Related Work

• The notion of order preserving encryption (OPE) [15–18] is another important encryption method in database to achieve the confidentiality while keeping the order of underlying plaintexts.
• Such a property allows users to perform comparison and range query over encrypted data without decrypting them.
• Another notion related is searchable encryption (SE) [19,20], which provides functionalities to perform keyword search over encrypted data without decrypting them.
• There are also some other related privacy-preserving methods proposed for the security in database [21–23,14].

3.1 System Model

• To provide data privacy protection solution and save upgrade cost for existing DB applications, L-EncDB system utilizes FPE technique to encrypt data.
• There are two layers in the L-EncDB system, that is, the application system layer and database layer.
• As shown in Fig. 2, the application system receives the interpreted SQL sentence by calling SQL interface, which takes the original SQL sentence as the input.
• The authors show how to process the interpretation for each type in their system.
• For fuzzy query or range query, SQL interpretation interface will further generate the query for ciphertext by using FQE or OPE and store it into an additional filed.

3.2 Security Notions

• For L-EncDB system, the interface for SQL interpretation can be deployed at client side or the application service layer.
• The authors assume that there exist authentication and access control methods to protect the key used in L-EncDB.
• The authors consider two types of attackers for L-EncDB: (1) attackers with access to database, including DBA or cloud service provider.
• They have access to the encrypted data and DB structure; (2) attackers with access to both application system and database.
• In another word, they are able to access SQL interpretation interface deployed in DB applications, construct SQL sentences with plaintext, gain interpreted SQL sentences with encrypted data, and view all fields and structure of database.

3.3 SQL-based Data Operations

• L-EncDB uses SQL interpretation interface to interpret all SQL sentence, which is viewed as an API that can be flexibly used by developers.
• Next, the authors describe the interpretation processes in details and show how L-EncDB supports SQL operations and queries over encrypted data.

3.3.1 Basic SQL Operations

• For basic data operation (insert, update, and delete) and query (exact query, join query, and nested query), the interface for SQL interpretation will replace the plaintexts with the corresponding ciphertexts encrypted by FPE.
• Note that FPE is deterministic, which means that SQL-based data operation and query operation can be directly executed over the encrypted database.
• “Insert into Table1(Field1, Field2) values (String1, String2 )”, will be interpreted to “Insert into Table1(Field1, Field2) values (fpek (String1 ), fpek(String2 ))”, where fpe is the adopted FPE algorithm, k is the selected encryption key, and fpek(x) means to encrypt x with fpe.
• Similarly, the interface for SQL interpretation will replace the constants in the queries with ciphertexts of FPE in the following SQL sentences: Update, “Update Table1 set Field2=String3 where Field1=String1”; Delete, “Delete from Table1 where Field1 = String1”; Exact query, “Select * from Table1 where Field1 = String1”; Join query, “Select Table1.* from Table1,Table2 where Table1.

3.3.2 Advanced Queries

• For range query, the interface applies OPE to generate ciphertexts and store them in additional fields.
• For a SQL sentence like “Insert into Table1(Field1) values (String1 )”, where Field1 is for range query, the interpreted SQL sentence will be “Insert into Table1(Field1, Field1Extra) values (fpek(String1 ),OPEk(String1 ))”, where Field1Extra is additional field for Field1, OPE is adopted OPE scheme and OPEk(x) means to encrypt x with OPE scheme using encryption key k.
• To perform fuzzy query, the interface changes the query filed to its additional field and generate keyword ciphertexts.
• Direct SQL-based operations cannot be supported in CryptDB, where the trusted database proxy is used to intercept all the SQL queries and decrypt their results.
• In short, compared with the other database encryption solutions such as CryptDB, L-EncDB is lightweight to support SQL-based operation directly in DBMS and can be flexibly deployed in database applications.

4 New FPE Scheme for Character String

• As described in section 2.1, there is no suitable FPE scheme for varchar data type with the restriction that the ciphertext has the same length with its corresponding plaintext.
• The authors propose a new FPE scheme for character string with arbitrary data type, which will be used in L-EncDB.

4.1 Preliminary

• Throughout the rest of the paper, the authors let Chars be a multi-byte character set, and Chars∗ be character strings over Chars of any length.
• For a multi-byte character set Chars, it can be divided into subsets and each subset contains characters of same size.
• The authors now give a review of the classical definition for FPE given by Morris et al. [2].
• (1) If a field is defined as nvarchar(n), it means that the field can store arbitrary character string with length (or character number) not more than n.
• The FFX method [3] is suitable for this type of domain.

4.3.1 Basic Idea

• For each character c, let v(c) be its mapping value in Zn.
• For strings with type of varchar, FPE will preserve both length and storage size.
• In the new FPE scheme, the result of addition has the same radix as that of left operand because the modulo is radix of left operand.

4.3.2 Description

• The authors new FPE can be described by three algorithms, that is, Setup, Encrypt, and Decrypt.
• It takes as input the string X, key k and the round number rn.
• More specifically, three steps will be included.
• Fig. 4 also describes the multi-radix modular addition algorithm, which ensures Ω(Li+1) = Ω(Ri) and Ω(Xi+1) = Ω(Xi), i.e., the output of i-th round has the same format as the input.

4.4 Security analysis

• According to [2], the PRP security notion under chosen plaintext attack, i.e., PRP-CPA, is defined by PRP game PRPAξ between challenger C and adversary A as follows.
• An FPE scheme ξ is PRP-CPA secure if any polynomial time adversary has only a negligible advantage in PRP game shown above, where the advantage is defined as AdvPRPξ (A) def = Pr[PRPAξ ⇒ 1]− 12 Theorem 1.
• If the underlying round function is a secure pseudo random permutation, their new FPE scheme achieves the PRP security.
• Assume that there exists an adversary breaks the security of their FPE scheme, a simulator will be built to show the insecurity of PRF , that is, the simulator could break the indistinguishability of PRF from a truly random permutation P .

5.1 Implementation Details

• The experiment for their L-EncDB system is conducted to evaluate its efficiency.
• To implement FPE schemes, AES and big number in open source library polarssl are used.
• That is, L-EncDB DLL must know DB structure, and the fields should be encrypted for fuzzy query or range query.
• Firstly, it analyzes SQL sentence, decides operations and tables to execute.
• Secondly, based on operation and table structure, it decides whether encryption, fuzzy query or range query are needed.

5.2 Experimental Evaluation

• L-EncDB is external encryption mechanism independent from database.
• To evaluate its performance, two issues are addressed: 1) the performance of FPE encryption algorithms for batch data encryption, 2) the performance of OPE.
• To construct the test platform, the authors use Java language to construct an application, in which the open source library named “HTTPClient” is used to visit the specified webpage in their website.
• As shown in Fig. 7.(a), the average execution time of query is about 0.6ms, but that of insert and update is only about 0.3ms.
• Execution time of SQL insert operation using L-EncDB is very close to that without encryption as shown in Fig. 7.(b).

6 Extension of L-EncDB to NoSQL Database Encryption

• With the advent of the Internet Web2.0 site, traditional SQL-based relational databases cannot be applied and many challenges arise, including: (1) Performance problem.
• In big data background, each user will produce massive dynamic information.
• BigTable of Goole and Dynamo of Amazon are successful implementation of NoSQL.
• In L-EncDB, exact query can be reserved because of the deterministic encryption.
• So, the fuzzy query expression can be interpreted as “db.users.find(“name” :/ˆFQEk(A)/)”.

Figures

Fig. 1. CryptDB architecture

Fig. 2. L-EncDB Architecture

Fig. 6. Execution time of FPE schemes

Fig. 3. Interpreting SQL sentences

Fig. 7. Comparison between SQL operations and encryption

Fig. 2. L-EncDB Architecture

Fig. 1. CryptDB architecture

Fig. 5. Implementation details

Table 1 FPE for data types in DB

Table 3 Execution time of encryption

Fig. 3. Interpreting SQL sentences

Fig. 4. FPE for character string and modular operations

Table 1 FPE for data types in DB

Fig. 4. FPE for character string and modular operations

Fig. 5. Implementation details

Fig. 7. Comparison between SQL operations and encryption

Table 3 Execution time of encryption

Fig. 6. Execution time of FPE schemes

Full text

L-EncDB: A Lightweight Framework for
Privacy-Preserving Data Queries in Cloud
Computing
Jin Li
a,∗
a
School of Computer Science and Educational Software
Guangzhou University, Guangzhou, 510006 P.R. China
Zheli Liu
b,∗
b
College of Information Technical Science
Nankai University, P.R. China
Xiaofeng Chen
c
c
State Key Laboratory of Integrated Service Networks (ISN)
Xidian University, Xi’an, P.R. China
Fatos Xhafa
d
d
Department of Languages and Informatics Systems
Technical University of Catalonia, Spain
Xiao Tan, Duncan S. Wong
e
e
Department of Computer Science
City University of Hong Kong, Hong Kong
Abstract
With the advent of cloud computing, individuals and organizations have become in-
terested in moving their databases from local to remote cloud servers. However, data
owners and cloud service providers are not in the same trusted domain in practice.
For the protection of data privacy, sensitive data usually have to be encrypted be-
fore outsourcing, which makes effective database utilization a very challenging task.
To address this challenge, in this paper, we propose L-EncDB, a novel lightweight
encryption mechanism for database, which i) keeps the database structure, and ii)
supports efficient SQL-based queries. To achieve this goal, a new format-preserving
encryption (FPE) scheme is constructed in this paper, which can be used to encrypt
all types of character strings stored in database. Extensive analysis demonstrates
Preprint submitted to Knowledge-Based Systems 25 March 2014

that the proposed L-EncDB scheme is highly efficient and provably secure under
existing security model.
Key words: Data query, outsourcing, privacy, format-preserving encryption
1 Introduction
The ever-increasing amount of valuable digital data both at home and in
business needs to be protected, since its irrevocable loss is unacceptable. The
advent of cloud storage motivates enterprises and organizations to outsource
data storage to third-party cloud providers, as evidenced by many real-life
case studies. Cloud storage services promise to be a solution for this problem.
In recent years, their popularity has increased dramatically. They offer user-
friendly, easily accessible and costsaving ways to store and automatically back
up arbitrary data, as well as data sharing between users and synchronization
of multiple devices.
As in any existing application and system, security and privacy play an ex-
tremely important role for the success, and certainly raise new challenges
among the many others that cloud storage is confronted with. Specifically,
when entrusting data to the cloud, data owner also releases control over the
data, resulting that their trust is put in the cloud service provider’s integrity
and in the security of its access control mechanisms. However, individuals and
especially businesses hesitate to entrust their data to cloud storage services
since they fear that they will lose control over it. Recent successful attacks
on cloud storage providers have exacerbated these concerns. The providers
are trying to alleviate the situation and have taken measures to keep their
customers’ data secure. The simple and popular solutions adopted for data
privacy are traditional encryption techniques such as public key encryption
or symmetric key encryption. Through these encryption methods before out-
sourcing, the security of users’s data can be protected.
However, traditional database encryption will change the data structure of
original data, and results in the impracticability of database application for
various kinds of SQL operations. If the data structure is changed, it cannot
support data operations over ciphertext such as range query and fuzzy query,
etc. Especially, there has been considerable recent interest in the paradigm of
∗
Corresponding author. Jin Li and Zheli Liu contribute to this work equally.
Email addresses: jinli71@gmail.com (Jin Li ), liuzheli1978@163.com (Zheli
Liu).
2

data mining-as-service: a company (data owner) lacking in expertise or com-
putational resources can outsource its mining needs to a third party service
provider. However, both the outsourced database and the knowledge extracted
through data mining are considered private property of the data owner in many
applications. Thus, to protect data privacy while realizing data mining and
knowledge extraction, the data owner is required to transform its data without
changing its structure before outsourcing.
Contribution. To realize effective data utilization after secure outsourcing,
we propose a lightweight encrypted database mechanism denoted by L-EncDB.
This new mechanism is able to protect sensitive information while keeping the
data structure in outsourcing service for big database application. In the pro-
posed L-EncDB system, the encryption and query are based on SQL. Through
only one interface, all SQL sentences for database can be interpreted.
Furthermore, based on format-preserving encryption (FPE) technique and a
new character string FPE scheme, L-EncDB can be implemented to preserve
data type and length in ciphertext. It enables i) to encrypt data and store
them without changing original database structure, ii) to perform SQL oper-
ations on all kinds of databases, including text database such as SQLite and
Access, iii) to support SQL-based operations including advanced fuzzy and
range queries.
Innovation. In this paper, a novel FPE scheme with the method of “multi-
radix modular addition” is proposed to support the L-EncDB lightweight
framework for privacy-preserving outsourced database. The new proposed
FPE can preserve both length and storage size of character strings, which
cannot be efficiently achieved in the traditional FPE schemes. Based on the
FPE scheme, data operations such as data mining and SQL-based queries can
be directly executed over ciphertexts in the proposed L-EncDB framework.
Furthermore, L-EncDB framework can be extended to text database (such as
SQLite used in mobile) and NoSQL databases, which have not been considered
in the previous related work.
1.1 Organization
The rest of this paper proceeds as follows. In Section 2, we give a survey for
the related work to ours. In Section 3, we propose the system architecture and
construction method for the L-EncDB system. In Section 4, we propose a prac-
tical construction of FPE for character string. Its security and performance
analysis is also given in this section. In Section 5, we present the implementa-
tion of prototype for L-EncDB with the proposed FPE, and in Section 6, we
present an extension of the L-EncDB to NoSQL database encryption. Finally
3

we draw conclusion and show the future work in Section 7.
2 Related Work
We briefly discuss FPE technique and privacy-preserving database encryption
solutions in this Section.
2.1 FPE
The notion of FPE [1–4] has been proposed to generate ciphertext with the
same format as plaintext while encrypting sensitive information. More specifi-
cally, FPE can keep data type and length in the ciphertext, therefore, without
changing database structure and field type. Thus, the use of FPE enables
upgrading database security in a transparent way. The goal of FPE is to
generate ciphertext which falls in the same domain as the plaintext. Some
practical FPE schemes have been proposed for simple domains such as integer
[5], character data [3] and datetime [6]. Character data is the common data
type in database, which appears in the form of character strings, i.e., the fi-
nite sequences of characters from some character sets. However, there is no
suitable character FPE solution to preserve both length and storage size of
strings above. For a string with character from iso-8859-1 or ASCII, where
the storage size of each character is 1 byte, the length of string is equal to its
storage size, and FFX is also suitable in this case.
However, most of character sets are represented using more powerful encoding
formats, and different characters may require different byte counts to repre-
sent. In this paper, such a character set is called “multi-byte character set”.
Consider a character string of length n with each character in UTF-8, its stor-
age size will be from n bytes to 4n bytes. In this case, FFX is unsuitable. In
2012, Li et al. [7] proposed a solution based on cycle-walking [8]. However,
cycle-walking can not guarantee stable efficiency, which is impractical in most
applications. In this paper, we develop a new FPE method in Section 4.
2.2 Privacy-preserving Database Encryption
A number of research results [9–13] were proposed for privacy-preserving
database encryption. However, most of them cannot provide the complete
solution for general SQL-base operations over encrypted data. To support
query over encrypted numerical data, Hakan et al. [9] firstly presented a basic
4

Fig. 1. CryptDB architecture
Database Layer
Field_1
Field_n Additional Field
ciphertext ciphertext Keyword ciphertext
Additional Field
Keyword ciphertext
Application
System
Trusted SQL Interpretation Layer
Format-preserving
Encryption
Fuzzy Query
Encryption
Order-preserving
Encryption
1. Plaintext SQL
2. Ciphertext SQL
3. Execute SQL
SQL Flow
Data Flow
Table1
Fig. 2. L-EncDB Architecture
framework of how to ensure data security in “Database-As-Service” (DAS)
model, in which a coarse query is executed by the database service provider.
Based on this basic framework, Wu et al. [12] described a solution for query
over encrypted character strings.
One of the most typical database encryption solutions is CryptDB [14], which
explores an intermediate design point to provide confidentiality for applica-
tions that use database management systems (DBMSes). As shown in Fig. 1,
CryptDB works by intercepting all SQL queries in a database proxy, which
rewrites queries to execute over encrypted data (CryptDB assumes that all
queries go through the proxy). The proxy encrypts and decrypts all the data,
and changes some query operators, while preserving the semantics of the query.
However, CryptDB is not designed for existing database applications and the
DAS model of cloud storage. In cloud computing, users are able to store,
modify and retrieve data from anywhere in the world, as long as they have
access to the Internet. CryptDB changes the database structure and stores
the ciphertexts generating by different encryption methods.
2.3 Other Related Work
The notion of order preserving encryption (OPE) [15–18] is another important
encryption method in database to achieve the confidentiality while keeping
the order of underlying plaintexts. Such a property allows users to perform
5

Frequently asked questions

Q1. What have the authors contributed in "L-encdb: a lightweight framework for privacy-preserving data queries in cloud computing" ?

To address this challenge, in this paper, the authors propose L-EncDB, a novel lightweight encryption mechanism for database, which i ) keeps the database structure, and ii ) supports efficient SQL-based queries. To achieve this goal, a new format-preserving encryption ( FPE ) scheme is constructed in this paper, which can be used to encrypt all types of character strings stored in database. 

Q2. What have the authors stated for future works in "L-encdb: a lightweight framework for privacy-preserving data queries in cloud computing" ?

Finally, the authors also showed how to extend their L-EncDB to realize the privacy-preserving queries over encrypted NoSQL Database. Especially, the authors will extend it to privacy-preserving knowledge extraction for outsourcing database, and further provide some practical data publishing methods suitable for their framework. Finally, the authors also showed how to extend their L-EncDB to realize the privacy-preserving queries over encrypted NoSQL Database. Especially, the authors will extend it to privacy-preserving knowledge extraction for outsourcing database, and further provide some practical data publishing methods suitable for their framework. 

Q3. How long does it take to execute a SQL insert sentence?

For a SQL insert sentence with 30 fields and no range query or fuzzy query, the execution time for each insert operation is about 0.09ms, i.e., the system can interpret 11000 SQL sentences within 1s. 

Q4. How do the authors construct a test platform?

To construct the test platform, the authors use Java language to construct an application, in which the open source library named “HTTPClient” is used to visit the specified webpage in their website. 

Q5. What are the popular solutions for data privacy?

The simple and popular solutions adopted for data privacy are traditional encryption techniques such as public key encryption or symmetric key encryption. 

Q6. What is the purpose of the cloud computing?

In cloud computing, users are able to store, modify and retrieve data from anywhere in the world, as long as they have access to the Internet. 

Q7. How many times are the user requests produced in web2.0 site?

To provide dynamic pages and information for user, thousands of read and write requests are produced each second in web2.0 site, and results in high concurrent load of database. 

Q8. What is the way to represent character data in a database?

most of character sets are represented using more powerful encoding formats, and different characters may require different byte counts to represent. 

Q9. What is the advantage of L-EncDB for existing applications?

for existing database applications, L-EncDB is more suitable and lightweight when enterprises and organizations outsource data storage to third-party cloud providers. 

Q10. What is the definition of a format-preserving encryption scheme?

A format-preserving encryption scheme is a function F : K × N ×T ×X → X ×{⊥}, where ⊥ ≠ X , and nonempty sets K, N , T , X denote the key space, format space, tweak space and domain, respectively. 

Q11. How can the proposed FPE be preserved?

The new proposed FPE can preserve both length and storage size of character strings, which cannot be efficiently achieved in the traditional FPE schemes. 

Q12. What are the advantages of using NoSQL in the cloud computing?

In the cloud computing, more and more enterprises also need to outsource NoSQL database to construct their business applications. 

Q13. What is the format space of a string?

In their FPE for strings with type of varchar, the format space is defined by both length and storage size, that is N [Chars] = {(l(X), s(X))|X ∈ X [Chars]}. 

Q14. Why do individuals and especially businesses hesitate to entrust their data to cloud storage services?

individuals and especially businesses hesitate to entrust their data to cloud storage services since they fear that they will lose control over it. 

Q15. What are the main reasons why L-EncDB is used in BigTable?

Experimental results demonstrate that the new L-EncDB is efficient and can be applied to big database for privacy-preserving applications. 

Q16. What is the syntax for a ciphertext in a SQL query?

As shown in Fig. 3.(a), for SQL data operation sentences, such as to insert, or to update, each constant in the query will be encrypted using FPE.