Skip to content

perf(vx-dte): stretch each platform's work by its own slowdown - #5305

Merged
Exelord merged 1 commit into
mainfrom
claude/project-thread-vn5f7g
Nov 13, 2026
Merged

Exelord merged 1 commit into
mainfrom
claude/project-thread-vn5f7g

Conversation

@Exelord

@Exelord Exelord commented Nov 13, 2026

Copy link
Copy Markdown
Member

Why

Main 5ba7a05: Mac tasks took 1.4x their history, Linux ones 1.15x. The plan stretched both by the run's 1.2x, so the Mac work looked to have room. Two shards waited 8.5 s for a core beside an idle Mac, and the Macs ended the run.

What

slowdowns() gives a slowdown per platform: a platform with 20 s of finished history gets its own, the rest keep the run's. The plugin and the simulator both stretch a task by the platform it runs on.

Proof

  • New sim test: 16 Linux tasks at 1.1x, 5 Mac tasks at 1.5x. Before, the Macs ended the run at 224 s; now it ends with the Linux work at 218 s. Fails without the change.
  • 75 seeded runs over three main profiles: -2.5 +/- 1.0 s.

Generated by Claude Code

Main 5ba7a05: Mac tasks took 1.4x their history, Linux ones 1.15x. The
plan stretched both by the run's 1.2x, so the Mac work looked to have
room: two shards waited 8.5 s for a core beside an idle Mac, and the
Macs ended the run. A platform with 20 s of finished history is now
stretched by its own slowdown; the rest keep the run's.

75 seeded runs over three main profiles: -2.5 +/- 1.0 s.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01W6xuYXmaeAXEqaqA1Z6RsZ
@Exelord
Exelord merged commit b6fcdde into main Nov 13, 2026
13 of 14 checks passed
@Exelord
Exelord deleted the claude/project-thread-vn5f7g branch November 13, 2026 13:41
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants