| Note | ||
|---|---|---|
| ||
This document is currently being drafted. Items on this page are expected to be in the release, but this is not guaranteed. |
What is this document?
The VIVO 1.8.1 release concentrates on fixes for certain bugs and performance issues. Some minor, non-breaking additions are present in the ontology, and a non-breaking addition to the UI.
Performance Improvements
| Center | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
Relative execution time. Testing by Jim Blake, as reported on the development list using the 30 slowest profiles.
|
Note that all testing reported below has been performed on a Macbook Pro, with PCI-E SSD. However, no specific tuning has been applied to either the hardware or software. Real world performance will depend on hardware and software configuration - it is recommended that you have an SSD / high IO performance storage layer, and if using SDB/MySQL, enough memory allocated to read the tables indexes.
Page Rendering
| Note |
|---|
AntiSamy is no longer used to filter fields before they are rendered. For a large profile in a test dataset, this was responsible for over two seconds of the execution time required to render a profile. A simple regular expression is used to filter out and JavaScript elements - this is 300x faster than using AntiSamy. |
...
- Reduce overhead on reinferencing - up to 30% faster on individuals with no changes (depending on configuration)
- Reasoning on a large dataset has more consistent performance (used to slow down / crash with memory used by the search indexer)
- Faux property resolution rewritten to greatly reduce work being repeated in the presence of multiple instances of the same property
- RDFService has additional methods
- CONSTRUCT that takes a Model to write into
- SELECT that takes a ResultSetConsumer - implemented by the user - which processes each QuerySolution as it is retrieved from the ResultSet
- Reduce latency and memory overhead of reading into a Jena model; serialising; and then re-reading into a Jena model in the calling method.
(NB: Responsible for 20 seconds of improvement to Map of Science / Temporal Graph)
- Replace certain uses of RDFServiceDatasetGraph with RDFService (repeated calls to find() in RDFServiceDatasetGraph responsible for some overhead)
- RDFServiceSDB always constructs queries against the graph, and not the union model (simple optional queries much faster against the graph than the dataset)
- Clean up of many list view SPARQL queries, removing a few redundant patterns.
- Cache list of graph lists when using a SPARQL backend for faster page loads (4 second saving on Virtuoso / 25 million triples)
- NOTE: Some methods have changed their signatures to support the above. If you have custom Java code in your installation, you may need to make minor adjustments - typically, this will be exchanging a Dataset parameter for an RDFService.
- NOTE: Some listview-*.xml files have changed, if you have customised your list views, you will need to resolve the conflicts.
- NOTE List views that return publications (e.g. authorInAuthorship) now only resolve the editor person for publications that are either bibo:Book or bibo:BookSection (includes Chapter, etc.). This is necessary for reasonable performance when you have large publication lists that involve articles with many co-authors.
...