Databases Reference
In-Depth Information
or on-the-fly recovery, new DQR theories are necessary to understand the
strength and weakness of existing DQR schemes, inspire the development of
novel DQR schemes, and make DQR a rigor field of study, for example. (2)
There is still a big gap in engineering practical DQR capabilities for real
world applications. For one example, Web Services (WS) and service-oriented
architectures have significantly changed the way applications are developed,
but no WS aware techniques have yet been developed to do transparent DQR
for WS-based applications. For another example, existing transaction-level
DQR mechanisms either require major changes in system design or suffer
from significant DoS or performance overhead.
Therefore, to fully solve the DQR problem, a holistic approach should be
taken to make an integrated set of innovative contributions on four funda-
mental aspects of DQR: theories, mechanisms, applications, and systems.
•
New DQR theories should be developed to (a) address quarantine and
transparency
, (b) define
quality
of DQR services, and (c) integrate
recov-
erability
and
quarantinability
.
•
New DQR schemes should be developed to advance the state-of-the-art
DQR techniques from the paradigm of read-write-dependency analysis to
the new paradigm of mark-based causality tracing, which will significantly
improve transparency and eciency.
•
Non-blocking repair schemes should be developed to advance the state-
of-the-art DQR techniques, from the paradigm of “clean-then-reexecute”
recovery to the new paradigm of “cleaning-free” recovery, which avoids the
overhead introduced by cleaning transactions.
•
New DQR schemes should be developed to advance the state-of-the-art
from the paradigm of “lock-competing reexecution” to the new paradigm
of “non-blocking repair”.
•
New DQR schemes should be developed to advance the state-of-the-
art from the paradigm of “pre-programmed DQR” to ”adaptive or self-
reconfigurable DQR”.
•
DQR theories and mechanisms should handle both conventional appli-
cations (which require ACID properties) and modern applications which
adopt a weaker consistency model to make distributed “business transac-
tion” processing (on top of Web Services) practical, scalable, and ecient.
•
From the perspective of system building, complete open-source DQR tools
and systems should be prototyped and evaluated using the appropriate
benchmarks.
Acknowledgement
Peng Liu was supported in part by NSF CCR-TC-0233324 and NSF/DHS
0335241.
