Time/Place

Attendees 

Agenda

  1. Voice F4 performance/scale interests and/or concerns
  2. Review previous F4 performance benchmarking (summary)
    1. Unimplemented "Technical Working Group" performance assessment plan
  3. Establish focus for next round of F4 performance benchmarking
  4. Establish collaborative benchmarking plan

Note: Once we have a new set of benchmarks, we can kickstart the follow-on effort of extending F4's scale.

Next call: 2015-11-09 Performance - Scale Meeting

Minutes

  1. Performance interests/concerns:
    1. Esme: want to get a good benchmark that is meaningful, can be used to test changes
      1. Also hearing concerns about Hydra/LDP/ORE proxies being chatty and slow, so want to look at client interaction model, look at both client and server improvements
    2. Danny: also interested in a good baseline to have solid data to work on
      1. Also interested in large binaries, like large video files
      2. As a vendor, interested in handling large datasets (several terabytes+), both in terms of large numbers of objects and large total size of preservation files
    3. Chuck: Concerned about scalability in total number of objects.  Have ~6 million objects now, anticipate 10-20 million.
      1. Transaction performance: detecting duplicates with Solr
        1. Danny: good point about the new stack: how do external services like Solr and triplestores fit into transactions, performance questions?
    4. Maurice: Mass-migration as content is moved into Fedora, performance of ingest/migration
      1. Also interested in repository performance during migration, and the tradeoff between that and ingest performance
      2. Moving towards more active resource management, want to verify resources as they move into Fedora
      3. Want to plan migration based on expected rate of ingest, want to know how to improve performance if rate of ingest isn't as good as we'd like
    5. Robert: Research datasets with Amazon infrastructure, interested in scalability in particular
    6. Nick: Interested in ingest performance esp. for migrating from Fedora 3, with fixity checking
      1. Large files (250GB video files)
      2. Clustering/sharding: need to figure out what role clustering can play and what scenarios it would help improve
  2. Process:
    1. Need to revisit benchmarks that we run to make sure they address the issues raised here
    2. Second agenda item has links to previous work
      1. There will need to be significant changes to revamp these to use a different testing tool (previous tool is no longer supported)
      2. Technical Working Group performance assessment plan divided up the testing space in slight different way
    3. Can use existing test results to identify particular performance concerns that seem more likely to be an issue versus those that look like they are performing better
    4. It would be good to re-verify some of the prior testing in a new framework that we can support moving forward
    5. Esme: good to review the test results to see if they address the concerns voiced here, and if so, if the results look promising or raise concerns
    6. Danny: the TWG work would be a good starting point to determine what kind of testing we should do
      1. Have used JMeter before – regardless of what tool is used, need to coordinate and share tools
    7. Nick: Sharing is key, would be good to setup an environment
    8. Esme: We could setup a Vagrant setup to make an easy-to-setup environment for consistent testing and lower barrier to performing testing
  3. Actions:
    1. Esme: review concerns raised today and determine how well it addresses concerns raised today
    2. Danny: look at creating sample dataset and specifying sample data (file sizes, number of objects, etc.)
    3. Maurice: help with specifying sample data
    4. Andrew: update wiki to organize ongoing and previous work
      1. Update community with today's discussion and next steps
  • No labels

All content on the LYRASIS Wiki is licensed under the CC BY (Attribution) license, unless otherwise noted.