Performance summary
https://wiki.duraspace.org/display/FF/Performance
https://wiki.duraspace.org/display/FF/Performance+Testing
- Defined testing profiles
- platform profiles
- workflow profiles
- repository profiles
- setup profiles
Single node
Properties CRUD
https://wiki.duraspace.org/display/FF/Fedora+4+Properties+CRUD+Performance+Testing
Objects CRUD
https://wiki.duraspace.org/display/FF/Fedora+4+Objects+CRUD+Performance+Testing
Large file
https://wiki.duraspace.org/display/FF/Large+File+Ingest+and+Retrieval
- Upload and Download as large as 1-TB via REST-API
- Likewise, using Federation
- Range retrieval verified on 1-TB on repo and federation
Large numbers
https://wiki.duraspace.org/display/FF/Large+Numbers+of+Objects
- Tested up to 10-millon objects
- Performance degrades linearly in proportion to:
- number of children of a resource
- number of resources in the repository
- Federation
- ~3000 files in a directory before performance degradation
- 16.7 million object in 4-level hierarchy out-performs 3-level directory hierarchy
Versioning
https://wiki.duraspace.org/display/FF/Versioning+Performance
- Up to 100,000 version of a single object created
- Timing collected for each new version creation
- Timing collected for creating new version of object with multiple datastreams
Flat hierarchy
https://wiki.duraspace.org/display/FF/Flat+Hierarchies+Testing
- Benchtool results are suspect
- Steady increase in response time as more children objects added to single parent
- Tests also included with transactions in use
Authz/no-Authz
- Read, update, delete tests performed
- Impact of AuthZ is 10% or less
F3/F4 comparison
https://wiki.duraspace.org/display/FF/Single-Node+Test+Results
Cluster
https://wiki.duraspace.org/display/FF/Response+Time+Comparison+of+Single+Fedora+VS+Cluster
- Load balancing over 3-F4 servers rotates serving requests as expected
- New servers can be added, existing servers can be removed
- Note: cluster was running on single machine