Performance Optimization

[wang1212]

In the process of upgrading G from v4 to v6, the underlying layer of G has been refactored a lot, from a single engine that only provides basic rendering capabilities to a plug-in engine, enriching a lot of ecological capabilities.

Unfortunately, performance issues were not paid close attention to during this process, resulting in a serious regression of performance benchmarks. So far, we have made a lot of efforts to optimize performance, but the results are still not ideal. The root cause is that our current strategy is relatively conservative. If we want to really improve performance, we need to make some adjustments to the current architecture. Considering the developer experience and maintenance costs, this is based on the premise that there are no major changes to the API.

Related performance issues:

• v5/v6 版本滚动性能下降明显 #1712

Here are some to-do items:

• Simplify event system design
  • Cancel unnecessary event bubbling perf: optimize element event distribution and attribute update process #1944
  • Refactor the event system to simplify logic complexity and event flow
• Simplify the element picking process in interactive events and remove rbush dependency
• Simplify hook design (redundant design leads to inconsistent plugin implementation, which increases complexity).
  • Simple and clear rendering process stage, commit to render
• Basic graphic elements
  • Measurements of text elements perf: use charWidthCache to improve performance by 1.5% #1978
• Benchmarks and Statistics
  • Comparison of performance benchmarks with other engines feat(benchmark): Add performance test suite and analysis panel #1987
  • Performance Testing in CI
  • Runtime performance statistics

Completed optimization (PRs):

• perf: optimize performance #1789
• perf: optimize large image rendering performance #1798
• perf: optimize canvas renderer performance #1810
• perf: use map instead of RefCountCache to avoid memory leaks #1939
• perf: discard animation when target is destroyed to avoid memory leaks  #1940
• perf: optimize element event distribution and attribute update process #1944
• perf: use charWidthCache to improve performance by 1.5% #1978
• feat(benchmark): Add performance test suite and analysis panel #1987
• perf: element event batch triggering #2005
• perf: remove rBush logic from element picking mechanism #2031

Reference performance benchmarks:

• https://benchmarks.slaylines.io/

Case	Before	After