FinTech and InsuranceTech case studies digitally transforming Europe's future with BigData and AI The new data-driven industrial revolution highlights the need for big data technologies to unlock the potential in various application domains. The insurance and finance services industry is rapidly transformed by data-intensive operations and applications. FinTech and InsuranceTech combine very large datasets from legacy banking systems with other data sources such as financial markets data, regulatory datasets, real-time retail transactions, and more, improving financial services and activities for customers.

1. Data Skipping technology
Yosef Moatti
IBM Haifa Research Labs
moatti@il.ibm.com
May the 14th – 20

2. BigDataStack & INFINITECH Webinar 2
1. Big (semi) structured data
2. Object Storage
3. Apache Spark: one of the most popular Analytics
engines for big data
4. More specifically Apache Spark SQL
Data Skipping: technology context
14.05.2020

3. Modern BigData architectures decouple storage from
compute
Hence:
Big data analytics implies a (big) data transfer
Therefore:
Data Skipping is important for BigDataStack
Proof: Data Skipping is already incorporated into IBM
products (see status slide)
BigDataStack & INFINITECH Webinar 3
Contribution to BigDataStack requirements
14.05.2020

4. BigDataStack & INFINITECH Webinar 4
1. Big (semi) structured data
2. Object Storage
3. Apache Spark: one of the most popular Analytics
engines for big data
4. More specifically Apache Spark SQL
Data Skipping goal is to
minimize the data transfer
Data transfer
Data Skipping: goal
14.05.2020

5. Example: Look for data in violent storm conditions
SELECT vessel_code, datetime, longitude, latitude, wind_speed
FROM cos://us-south/…/danaos stored as parquet
WHERE wind_speed > 30
Determine which objects are NOT relevant for a
SQL query using a data skipping index
Stores summary index metadata for each object
No change to Apache Spark base code
“just” take advantage of a Spark external API which
permits to refine the set of objects relevant to a given
query.
Skip over irrelevant objects
Skipped objects are not touched at all
IBM Data skipping is state of the art.
It comprises:
UDFs support
Data Skipping for LIKE and general regular expressions
Multiple indexes including Bloom Filter, etc…
Plug-in user indexes
Etc…
è Saves time and $
Index of All
Objects
QueryData Skipping
Index
Candidate
Objects
WHERE
Clause
Save Time
and $
Data Set Objects
Min/max index on
wind_speed column
Possibly relevant
object are ingested
Data Skipping: how does it work?
BigDataStack & INFINITECH Webinar14.05.2020

6. Data Skipping within the maritime use case
6
Danaos vessel dataset
Typical SQL queries: SELECT vessel_code, datetime, longitude, latitude, wind_speed
FROM cos://us-south/…/danaos stored as parquet WHERE wind_speed > 30
The Object Storage: either remote IBM Cloud Object Storage (COS) (as shown) or local
Depending on query and data partitioning we get up to 99% reduction in data transfer
BigDataStack & INFINITECH Webinar14.05.2020

7. Data Skipping for the BigDataStack maritime dataset demonstrated at THINK ’19
(joint work with Danaos: the BigDatastack maritime use-case partner)
Data Skipping: status
BigDataStack & INFINITECH Webinar14.05.2020
IBM Products
Database Catalog Support added to IBM Cloud SQL Query
Blog published here
Released as open beta
Data Skipping integrated within IBM Cloud SQL Query
Blog published here
Released as closed beta (as of now)
IBM Analytics Engine (IAE)
Data Skipping feature available as open beta
That’s all … for now
Scientific paper soon to be submitted

8. Data partitioning
Push for additional adoption
Decouple Data Skipping from Apache Spark
…
BigDataStack & INFINITECH Webinar 8
Data Skipping: next steps
14.05.2020

9. The team
Team:
Oshrit Feder
Guy Khazma
Gal Lushi
Yosef Moatti
Paula Ta-Shma
Technical contact point: PAULA@il.ibm.com
07.05.2020 BigDataStack & INFINITECH Webinar 9