Databases Reference
In-Depth Information
initial work [44, 26] in these areas, many extensions to the problem have been
considered. We briefly mention these advances that we have not covered so
far to provide interested readers with references.
Besides extending the data model, some researchers have considered relax-
ing assumptions made by the basic DAS model itself. The basic DAS model,
as discussed in this chapter, assumes “curious but honest” adversary, but such
an assumption might not necessarily hold in certain situations. In particular,
the service provider may return erroneous data. An error in the result to a
query may manifest itself in two ways - the returned answers may be tampered
by the service provider, or alternatively, the results returned by the service
provider may not be the complete set of matching records. The problem of in-
tegrity of the returned results was first studied in [26] for the untrusted server
model. Any authentication mechanism adds additional processing cost at the
client, and therefore authentication mechanisms using Merkle Hash trees and
group signatures that attempt to reduce such an overhead have been studied
in [38]. The authors have developed techniques for both the situation where
the client (i.e., the user who poses the query) is the same as well as different
from the data owner.
Another avenue of DAS research has been to exploit secure coprocessor
to maintain confidentiality of outsourced database [5]. Unlike the basic DAS
model in which the client is trusted and the service provider is entirely un-
trusted, in the model enhanced with a secure coprocessor, it is assumed that
the service provider has a tamper proof hardware - a secure coprocessor -
which is attached to the untrusted server and has (limited) amount of storage
and processing capabilities. Data while outside the secure processor must be
in the encrypted form, it could be in plaintext within the coprocessor without
jeopardizing data confidentiality. Exploiting a secure coprocessor significantly
simplifies the DAS model since now intermediate query results do not need
to be transmitted to the clients if further computation requires data to be
in plaintext. Instead, secure coprocessor can perform such a function, there-
fore significantly reducing network overheads and optimizing performance.
Another additional advantage is that such a model can naturally support sit-
uations where the owner of the database is different from the user who poses
the query. Another very similar approach using “smart cards” was proposed
in [9].
There are several interesting proposals for designing systems that support
querying and management of encrypted data [3, 9, 13]. [3] proposes a “two-
server” model where data vertical data partitioning and selective attribute
encryption is used for enabling confidentiality. [9] proposes an architecture
that uses a small trusted hardware (a “smart card”) to carry out computation
over plaintext data while the bulk storage and processing is carried out by the
untrusted server which has only access to the encrypted data. [13] propose
a secure B+-tree based indexing approach to query data kept on a single
untrusted server and analyze the disclosure risk in terms of inference-based
attacks where the adversary has different degrees of background knowledge.
