2016 11 14 Open NEST Developer Video Conference

previous | list | next

Agenda

1. Welcome

2. Report from NEST User Workshop 2016

   • very successful workshop
   • many new contacts between and among users and developers

3. Review of NEST User Mailing List

   Mails since last meeting were reviewed and handled. Some additional (meta-)questions came up:

   • at some point we need to consider opening a NEST on MAC team, also questions about python on mac came up.
   • what is the scope of nest-user? How far do we discuss neuroscience questions about the behavior of specific networks on the software mailing-list?
   • No answer on LayoutNetwork, so we take it as a GO and can change behavior in refactoring endeavors
   • How many more models to put into the kernel before moving to NESTML? And, at what point do we want to create a separate repository for user models?

4. Update: Work on GIDCollection and subnet removal

• See Project 2 and especially at the moment #455
• terhorstd, stinebuu, hakonsbm, heplesser working on it at present
• first major step is conversion of all functions (esp Create, Connect) to use GIDCollections
  • currently only primitive GIDCollections (composite being implemented)
  • requires adjustment of many tests and most examples
  • works partially at SLI level, Python interface being implemented
• When subnets disappear, also related functions such as GetChildren, GetLeaves, probably also GetNodes disappear.
  • A: no problem
• Main work branch currently heplesser/SubnetFreeNEST

2. Discussion: How to proceed towards NEST 2.12 and NEST 3

• Need to identify PRs and Issues to be included in NEST 2.12
• Need issue review session/team (info: new issue labeling scheme coming)
• Need to branch before integrating any subnet-removal-related material

• Decision: use the github milestones to sort the issues/PRs to NEST 2.12/3.0.

  @heplesser, @terhorstd and @jougs will form a task force for doing this.

2. Discussion: Removal of DataConnect(see #537)
   • need to rerun the benchmarks with more current code. @hbos is on it.
3. Discussion: LOG( M_ERROR, ...)
   • Shouldn't one just throw an exception instead?
   • Is this a left-over of old practice?
   • See conn_builder.cpp:589 for one of over 50 examples
   • A: probably a leftover, we can remove
4. #523 NEST connection infrastructure improvements: Presentation by Alexey Shusharin
   • Questions to answer:
     • what is the most common use-case for usage of connection
     • how important are cases with small number of connections (e.g. spike recorder/generator)
     • what is a use-case for target connection ports?
   • Q: if neuron receives inputs through static synapse, all inputs could be placed in one TargetConnector?
     • A: Yes, upto e.g. 64 connections of a single synapse model, then new TargetConnector. Other model would be another TargetConnector.
   • Q: How can we deliver spikes efficiently in a linked-list memory layout?
     • A: ConnectionHandle provides direct link to the target Connector.
   • C: very difficult to predict performance, because locality and cache performance are important. Can only find out by experimenting+benchmarking.
   • Q: how do you handle the sparse table?
     • A: The sparse table basically doesn't change, just the contained information is different.
   • C: bottleneck is the delivery and it's very difficult to predict the performance of the proposal due to interactions with ring-buffer entries.
     • going to post-synaptic storage would change a lot
     • proposal would probably be a good improvement (memory-wise) compared to 4g kernel
     • A: @alexeyshusharin wants to implement a prototype, then we can compare performance.
5. Review of open Github Pull Request
   • -will be handled in issue sorting task force-
6. Review of open Github Issues
   • -will be handled in issue sorting task force-