Download Overview

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Mobile device forensics wikipedia , lookup

Information privacy law wikipedia , lookup

Data remanence wikipedia , lookup

Transcript
Overview
PicoDBMS is a full-fledged DBMS (storage, indexation, query processing, access rights and
transaction control) embedded in a smart card. The application domain of PicoDBMS is the
management of shared secured portable folders (medical folder, user profile, agenda, etc.). Its
indexation and query engines implement new strategies required to cope with the smart card hardware
constraints. A first prototype written in JavaCard has been demonstrated at the VLDB’01 conference.
Since then, a second prototype has been written in C and optimized. At the same time, Schlumberger
provided us with an experimental smart card platform and modified part of its smart card OS, so that
the current prototype exhibits two order of magnitude better performance than its JavaCard
counterpart. Experimentations are still on going on this prototype.
Description
As smartcards become more and more versatile, multi-applications and powerful (32 bit processor,
more than 1MB of stable storage), the need for database techniques arises. Let us consider a health
card storing a complete medical folder including the holder’s doctors, blood type, allergies,
prescriptions, etc. The volume of data can be important and the queries fairly complex (select, join,
aggregate). Sophisticated access rights management using views and aggregate functions are required
to preserve the holder’s data privacy. Transaction atomicity and durability are also needed to enforce
data consistency. More generally, database management helps to separate data management code from
application code, thereby simplifying and making application code smaller. Supporting database
management on the card itself rather than on an external device is the only way to achieve very high
security, high availability (anywhere, anytime, on any terminal) and acceptable performance.
However, smartcards have severe hardware limitations which stem from the obvious constraints of
small size (to fit on a flexible plastic card and to increase hardware security) and low cost (to be sold
in large volumes). These limitations (tiny RAM, little stable storage, very costly write and lack of
autonomy) make traditional database techniques irrelevant. Typically, traditional DBMS exploit
significant amounts of RAM and use caching and asynchronous I/Os to reduce disk access overhead as
much as possible. With the extreme constraints of the smartcard, the major problem is scaling down
database techniques.
PicoDBMS design has been made with smartcard applications in mind but its scope extends as well to
any ultra-light computer device based on a secured monolithic chip.
PicoDBMS conception was guided by a set of design rules derived from the smart card properties:
1. Compactness rule: minimize the size of the data, indexation structures and PicoDBMS footprint
to cope with the limited stable storage area.
2. RAM rule: minimize the RAM consumption of all operators given its extremely limited size.
3. Write rule: minimize write operations given their dramatic cost (10 ms/word).
4. Read rule: take advantage of the fast read operations in stable storage (100 ns/word).
5. Access rule: take advantage of the low granularity and direct access capability of the stable
memory for both read and write operations.
6. Security rule: never externalize private data from the chip and minimize the algorithms
complexity to avoid security holes.
7. CPU rule: take advantage of the over-dimensioned CPU power, compared to the amount of
embedded data.
Following these rules, the technical solution relies on highly compact storage structures (attributes are
stored in domains, acting as a compression by dictionary), on ad-hoc compact indexation techniques
(attributes possibly engaged in selections and joins are linked by rings of pointers), and on a pure
pipeline query execution model consuming a minimal (bounded) amount of RAM.
Related papers
(les reprendres du site web pour avoir les liens)
N. Anciaux, L. Bouganim, P. Pucheral, ‘Data confidentiality: to which extent cryptography
and secured hardware can help’, Annals of telecom, to appear in 2005.
N. Anciaux, L. Bouganim, P. Pucheral, ‘Database Components on Chip’, ERCIM News, (54),
2003. (Article invité). pdf.
P. Pucheral, L. Bouganim, P. Valduriez, C. Bobineau, 'PicoDBMS: Scaling down Database
Techniques for the Smartcard', Very Large Data Bases Journal, VLDBJ, 10(2-3), 2001.
Special issue on the best papers from VLDB’2000. pdf.
N. Anciaux, L. Bouganim, P. Pucheral: ‘Memory Requirements for Query Execution in
Highly Constrained Devices’. 29th International Conference on Very Large Data Bases,
VLDB'03, September 2003.pdf.
N. Anciaux, C. Bobineau, L. Bouganim, P. Pucheral, P. Valduriez, 'PicoDBMS: Validation
and Experience'. 27th International Conference on Very Large Data Bases, VLDB'01,
September 2001. (demo session). pdf.
C. Bobineau, L. Bouganim, P. Pucheral, P. Valduriez, 'PicoDBMS: Scaling down Database
Techniques for the Smartcard', 26th International Conference on Very Large Data Bases,
VLDB'00, (10 pages), September 2000. (Best Paper Award). pdf.