query execution assurance for outsourced databases 1.31 - 08/29/05 prepared for vldb 2005 radu sion...

Query Execution Assurance for Outsourced Databases

1.31 - 08/29/05 prepared for VLDB 2005

Radu Sion([email protected])

http://www.cs.stonybrook.edu/~sion

Secure Systems Lab

Computer SciencesStony Brook University

2

Radu SionRadu Sion Securing Outsourced Data

outline

data outsourcing query completeness searching secure co-processor

prepared for VLDB, 8/05

3


data outsourcing


Data Server(Bob)

Data Client

(Alice)

Q1,…,Qb

Q1)…,(Qb)

1

2

OnlineQueryInterface

Outsourced Data

client• PDA• personal email user• file-client

server• email server• PostgreSQL• file-server

4


data outsourcing: challenges


Untrusted server: • lazy: incentives to perform less • curious: incentives to aquire information• malicious:

• denial of service • incorrect results• possibly compromised

Why is this hard ?• how ?• arbitrary expressivity• overheads

• network• computational costs

What do we do ?• query assurances• full privacy

• of queries (even encrypted)• of access patterns

• data confidentiality

5


pointer



6


querying with completeness: why ?!


Client requires quantifiable assurances that query results are complete and correct, for arbitrary query types in the presence of a server that could be …

… lazy (we do this *here*)

… and/or fully malicious (!)

7


ringers


P. Golle and I. Mironov,”Uncheatable Distributed Computations”, RSA 2001 (Cryptographer's track)

Alice Bob

a1,…,an1

f(a1),…,f(an)2

b1,…,bx,…,bnf(bx)1

f(b1),…,f(bn) x’ 2x’=x ?3

a1,…,an,f()

f(ai)=? f()

8


query completeness proofs (lazy server)


Database ServiceProvider (Bob)

Data

Data Owner(Alice)

S1

Si

Ss

x=x’?3

Q1,…,Qb,C(Q,x,)1

Q1)…,(Qb),x’2

Que

ryIn

terf

ace

A challenge token (computed by client) is sent together with the batch of queries. Upon return, batch execution is proved by (x==x’).

9


secure query interface (SQi)


network

sqi.server.QueryServer

DBMS

Database Service Provider

Traditional JDBC

sqi.client.QueryClient

Extended JDBC Interface

Traditional JDBCSecurity Controls

Secure Client

Data Client

securityevent

10


SQi: client interface


sqi.client.QueryClient

Legacy Query Interface Security Controls

Data Client

securityevent

Fast Crypto

PrivacyManager

Event Callback Management

Extended Query Interface

Security LogicExecution Proofs

Query Pre-processor

PersistentQuery Store

Transparent QueryBatch Scheduler

Client DataSource Adapter

to

server

Plugin Handler

11


SQi: server


sqi.server.QueryServer

Security LogicExecution Proofs

Server DataSource Adapter

Incoming Query Queue

QueryPost-processor

PrivacyManagerAgent

LegacyData

Source

Fast Crypto

from

client

Plugin Handler

12


success probability of cheating


13


execution times


14


overheads are reasonable


15


beyond laziness


• client-side result checking • weaker assurances of a stronger type

• secure hardware (we’ll see later)

• etc. ?

16


pointer



17


searching: fun for sure, but important?


Selected scenarios • compromised server (e.g. network context)• secure email server

• do not allow sysadmin to read email

• secure networked file system• unable to deploy forensics (without data owner consent)

• secures (from commercial competition):

• company data• data access patterns

sample: “return all emails containing ‘John’ and ‘lunch’”

18


searching: fun for sure, but important?!


Challenges • result assurances

• completeness• correctness

• confidentiality of data• obliviousness

• privacy of searches• no correlation leaks

• overheads • computational• network• storage constrained client

• dynamic (updates)

19


searching: helicopter overview


document serverdata client

…

…

k1

k2

k3 x

xx

d’4d’3d’2d’1

k4 x

C~

x

x

x

k1: (d1+x)(d3+x) mod p

k2: (d2+x)(d3+x) mod p

…

k1

k2

xx

d’4d’3d’2d’1C~

x

…

k1

k2

xx

d4d3d2d1C

xx

d1: k1

d2: k2

d3: k1 k2

d4: k3 k4

query: {k1 ,k2}1

F-1

4

retrieve: d3

6

qnr1,qr2, qr3,…, qrk

2

qr’1,qnr’2, qr’3,... , qr’k

v’1,... ,v’n

compute vi values3

v1,... ,vn

5

verifychecksums

Deploying a modified version of computational PIR targeted at a server-side indexing structure to achieve complete privacy.

Computational Privacy

QueryCompleteness

20


pointer



21


e.g. IBM 4758 (4764)


22


architecture overview (4758)


23


trust propagation (4758)


24


scpu: possible benefits


data management server

Server Storage

OutsourcedData

(encrypted)

Host CPUdata client

secure

insert/update

arbitrary

private query

encrypted query

response

SecureMemory

Secure Co-Processor

A secure co-processor on the data management side may allow for significant

leaps in expressivity for queries where privacy and completeness assurance are important.

encrypted item

25


scpu: searching with privacy


document server

Server Storage

OutsourcedDocuments(encrypted)

Host CPUdata client

conjunctive keyword

search query

encrypted query

response

SecureMemory

Secure Co-Processor

For conjunctive keyword searches on document (email, files) servers, oblivious search index

structures could be queried in secure memory achieving a novel zero-leak query model.

search index

secure

insert/remove

update index

26


scpu: hash-join (with privacy)


database server

Server Storage

OutsourcedRelations

(encrypted)

Host CPUdata client

P x Q

private query

encrypted query

response

SecureMemory

Secure Co-Processor

Hash-JOIN could be naturally accomodated.

P

Q

HP

27


scpu: merge-join (with privacy)


database server

Server Storage

OutsourcedRelations

(encrypted)

Host CPUdata client

P x Q

private query

encrypted query

response

SecureMemory

Secure Co-Processor

For Merge-JOIN, order-preserving encryption primitives could be deployed to minimize the amount

of data parsing required in the sorting phase.

P Q

28


scpu: what about general semantics ?


How do we aproach the problem of arbitrary query expressivity with strong computational (at least)

privacy ?

Let’s look at things we don’t “believe” in …

29


sample “wouldn’t do”: SCPU=client proxy


database server

Host CPUdata

client

queriesSecure Co-Processor

Server Storage

OutsourcedData

“client-server”interaction

“client proxy”

crypto work

database server

Host CPUdata

client

queries

Server Storage

OutsourcedData

client/serverinteraction

crypto work

crypto work

good idea ?not so sure !

30


scpu: some things we are afraid to do


• Process entire queries on SCPU (!)

• Dedicate (one) SCPU per query or equivalent• e.g., limit TPS by SCPU TPS

• Synchronize CPU with SCPU• e.g., block main CPU until SCPU completes

• Transfer >= O(n) on SCPU-CPU bus (!)

• Anything else un-smart

31


selected related research (SCPU)


Kenneth Goldman, Enriquillo Valdez: “Matchbox: Secure Data Sharing”, IEEE Internet Computing 2004

“Practical server privacy with secure coprocessors”, IBM Systems Journal 2001, S. W. Smith, D. Safford

J. Marchesini, S.W. Smith, “SHEMP: Secure Hardware Enhanced MyProxy”Technical Report TR2005-532, Department of Computer Science, Dartmouth College, February

2005.

A. Iliev, S.W. Smith, "Protecting Client Privacy with Trusted Computing at the Server", IEEE Security and Privacy, March/April 2005

A. Iliev, S.W. Smith, "Private Information Storage with Logarithmic-space Secure Hardware.", 3rd Working Conference on Privacy and Anonymity in Networked and Distributed Systems.

A. Iliev, S.W. Smith, "Prototyping an Armored Data Vault: Rights Management on Big Brother's Computer.", Privacy-Enhancing Technology 2002

E. Mykletun and G. Tsudik, “On using Secure Hardware in Outsourced Databases”, International Workshop on Innovative Architecture for Future Generation High-Performance

Processors and Systems, January 2005

Related research at IBM TJ Watson (Bishwaranjan Bhattacharjee a.o.)

32


cat /proc/lunchtime


Thank You

query execution assurance for outsourced databases 1.31 - 08/29/05 prepared for vldb 2005 radu sion...

Documents

outsourced data sqi

outsourced data pointer

outsourced data querying

outsourced data ringers

client interface

query results

query assurances

radu sion securing outsour