Queriable “Secondary Storage” Data

It is said that the primary use case for object stores is to serve as secondary storage. With the increasing amount of data being gathered and analysed (have someone said IoT?) much of this data will make it to secondary storage.

Being kept on secondary storage does not mean that the data does not need to be queriable anymore: A recent identified trend may be searched for in older data that was moved to secondary storage. Storlets allow an efficient and simple querying of data that resides in Swift.

Another closely related use case is that of aggregation. It is a well known practice to aggregate data as it gets older. Storlets can serve as ‘in place’ data aggregators.

Below are more concrete use cases that fall under the definition of efficient queriable secondary storage data.

Pushing down SQL filtering from Spark

Apache Spark is a most popular analytics engine that has multiple plugins for various types of analytics workloads. In addition Spark can work with various backend storage systems, with Swift being one of them. Spark SQL is a Spark plugin that allows to analyse structured data. At the heart of Spark SQL there is an SQL engine called “Catalyst”. Given an SQL query “Catalyst” identifies the filtering part of it. Thus, the filtering part can be pushed down to a Storlet. The idea was presented in the Tokyo Openstack summit, and can be viewed in [1].

Analysis over binary data

Analytics is typically done over textual data. In some cases that data is embedded in a binary format. Storlets can be used to extract the textual data from the binary object, thus saving the need to download it prior to extraction. One such example is exif metadata in jpegs. This ideas was presented in the Paris Openstack summit, and can be viewed in [2].

[1]https://www.youtube.com/watch?v=v9KCh–6Zw8
[2]https://www.youtube.com/watch?v=7tqMT0arV2k