Page tree
Skip to end of metadata
Go to start of metadata

Time/Place

Attendees 

Agenda

  1. Review last meeting actions: See actions section below
  2. Status of "many members" testing:  FCREPO-2105 - Getting issue details... STATUS
  3. Input on AWS Credits Proposal (guidelines)
  4. Documentation for running tests and creating graphs
  5. Define tests of known pain points, e.g.
    1. Performance of "many members"
    2. Ugliness of pairtree resources
    3. migration of PIDs from fcrepo3 to fcrepo4
    4. Migration from fcrepo3 to fcrepo4 in general and in a reasonable amount of time for 1M+ objects
    5. Loss of system dates, the migration of creation/modification times from fcrepo3 to fcrepo4
    6. Intra-repository referential integrity can be a pain point, but it has benefits too

Minutes

Regarding the performance test summary report:  Nick Ruest was waiting on Colin Gross  to get the R scripts together.    But Colin has generated the data.  Analysis and summary of graphs is needed.  We should come up with some questions contextualize the graphs?  For example:  How does performance change as resources increase?   Then we could drill down onto each kind of resource - size, type, etc.  Long term it would be helpful to create a template for the summary so that we can generate the summary automatically based on the summaries.

Questions / Advice for a State of Performance and Scale page: 

Joshua Westgard would be interested in best practices for performance wiki page :  

What are the different ways to set up a repository and what are the performance characteristics and performance mitigation strategies for each type.

How to protect against performance degradation as repository grows?

Esmé Cowles seconded that thought.


Colin Gross  is working on a mock file system that will enable testing of very large files.    The file system path of an item is based on the SHA-1.    The metadata cannot go into the mock system.   Colin Gross 's python based FUSE file system is not quite ready.  He will let us know when it reaches 0.1 he will let us know.

Esmé Cowles is has tested Danny's performance changes.  He has not seen the performance improvements.  But thinks it may have something to do with large numbers of hash-uris stored on a node and possibly inefficiencies in hydra updates.   Developing...


Actions

 

  • No labels