Performance Investigations

[Owen Arnold]

Aim here is three fold.

• Firstly, identify or generate some standard reductions scripts to use as benchmarks
• If this could be wrapped up in a performance test(s), that would be great.
• Compile a list of performance bottlenecks.

Only after this can we make targeted performance improvements. Each enhancement occurring as a new ticket.

1-Week effort

[Nick Draper]

Bulk move of tickets out of triage (new) to assigned at the introduction of the triage state

[Owen Arnold]

1) Jean is performing the critical benchmarking for the SNS #8929

2) The new c++ workflow algorithms are much faster than their older python counterparts. This has been reported by the users, who are now happy with the speed. The main bottleneck is in IO file loading. There are two aspects here.

• We have reduced the loading of files in the GUI. For example, we attempt to extract run numbers from titles where possible. #9167
• There is an already a ticket to look at HDF5 instead of Nexus APIs for IO. This is the only sensible route to IO speedup. #9450

[Owen Arnold]

No code changes made.

[Karl Palmen]

This ticket is a blocker, but the ticket #8929 for the benchmarking is not a blocker. This ticket would pass if the priorities were changed appropriately.

[Owen Arnold]

Replying to Karl Palmen:

This ticket is a blocker, but the ticket #8929 for the benchmarking is not a blocker. This ticket would pass if the priorities were changed appropriately.

This ticket is not blocking anything. I'm putting it back into the pool.

[Stuart Campbell]

This ticket has been transferred to github issue ​9787