expressive power comes great challenges.” In this article, we have highlighted a few recent successes in tackling these challenges, but there remain rich opportunities for further contributions to the field. Productive future work areas include extending the holistic optimization concept to new domains (for example, machine learning), and leveraging query and data characteristics to deliver tighter robustness guarantees.

References

1. Ashoke, S. and Haritsa, J. CODD: A dataless approach to big data testing.

PVLDB 8, 12 (Aug. 2015), 2008–2011.

2. Chandra, B., Chawda, B., Kar, B., Reddy, K., Shah, S. and Sudarshan, S. Data generation for testing and grading SQL queries. VLDB J. 24, 6 (Dec. 2015),

731–755.

3. Dutt, A. and Haritsa, J. Plan bouquets: A fragrant approach to robust query processing. ACM Trans. Database Syst.

41, 2 (June 2016), 11–1: 37.

4. Emani, K. V., and Sudarshan, S.

Cobra: A framework for cost-based rewriting of database applications. In

Proceedings of the IEEE Intl. Conf. on Data Engg. (Apr. 2018), 689–700.

5. Karthik, S., Haritsa, J., Kenkre, S., Pandit, V. and Krishnan, L. Platform-independent robust query processing.

IEEE Trans. Knowl. Data Eng. 31, 1

(Jan. 2019), 17–31.

6. Ramachandra, K., Chavan, M., Guravannavar, R. and Sudarshan, S. Program transformations for asynchronous and batched query submission. IEEE Trans. Knowl. Data Engg. 27, 2 (Feb. 2015), 531–544.

Jayant R. Haritsa ( [e-mail]) is a professor at the Indian Institute of Science, Bangalore, India.

S. Sudarshan ( [e-mail]) is a professor at the Indian Institute of Technology, Bombay, India.

© 2019 ACM 0001-0782/19/11

to speed up data access.

The transformations successfully carry out batching and asynchronous submission of queries, 6 prefetching of query results, and conversion of procedural code to SQL.

A metaphorical depiction of batching rewrites in DBridge is shown in Figure

2, where queries that are issued one-at-a-time, symbolized by the individual “taxis,” are batched into a single unified request, carried by a “bus.” Each transformation caters to a restricted scope and is therefore easy to prove correct, but in tandem they can successfully rewrite complex application programs. Further, the Cobra component of DBridge4 efficiently chooses the least cost program from many alternative transformed programs, by leveraging concepts from query optimization based on algebraic equivalence rules.

Techniques for holistic
optimization of queries con-
taining imperatively coded
user-defined functions
(UDFs) were developed
jointly by IIT Hyderabad
and IIT Bombay; some of
these mechanisms have
subsequently been imple-
mented and released in
Microsoft SQL Server 2019,
garnering excellent reviews
from users.c
Collectively, the DBridge
suite of techniques brings
the powerful benefits of de-
clarative query optimization
to imperative code, open-
ing a new research frontier.
More details on these tech-
niques may be found on the
DBridge project home page.

Query and Engine Testing
With the onset of the Big
Data world, where data is
the engine driving virtu-
ally all aspects of human
endeavor, it is vitally
important to ensure both
the applications and the
underlying platforms are
functionally correct. The
XData systemd developed at
IIT Bombay supports test-
ing of SQL queries by gen-
erating datasets designed
to detect many types of
common errors. 2 XData can
be used in database courses
to help students master
the nuances of SQL query
formulation and verify their
correctness; further, the
XData system facilitates au-
tomated grading of incorrect
queries by assigning partial
markings that reflect the
severity of the errors. XData
c https://www.microsoft.com/en-us/
research/project/froid
d https://www.cse.iitb.ac.in/infolab/
xdata
is currently operational at
multiple universities.

The testing of Big Data platforms is addressed by the CODD projecte at IISc Bangalore, using a distinctive metaphor of “dataless databases.” 1 Here, databases with a desired set of characteristics can be efficiently simulated without explicit creation or persistent storage of the contents. This approach is essential since traditional testing techniques, which involve construction of representative databases and regression query suites, are completely impractical at Big Data scale due to the time and space overheads involved in their execution. The CODD tool has been successfully used for testing of database engines in the software and telecom industries.

Future Research

An important reason for the
rapid adoption of SQL in
the 1970s was its simplicity,
which lent itself to effective
query optimization. Howev-
er, a host of complex features
have been added over the
years, and today’s query
processing world can be
paraphrased as “with great
e https://www.cse.iitb.ac.in/infolab/
xdata

P er for man c eD eg ra datio n

108

107

106

105

104

103

102

101

100

3D_ Q15

3D_ Q

96

4 D_Q

7 4D_ Q 26

4 D_Q 2 7

4 D_Q 91

5

D_Q1

9

5D_Q

2

9

5D_ Q84

5D_ Q18

Native Optimizer Spillbound

Figure 1. Performance robustness profile.
Original
Application
Program
Database
When will
my turn come? Queue of
Queries
Optimized Database Access:
Faster, Cheaper

DBridge Program Rewrite System

Rewritten
Application
Program
Database

Figure 2. Rewrites for optimizing data access.