Repository navigation
Replies: 2 comments
Correct
Yes, you've selected a parallel simplex algorithm, so the
For simplex we decided not to switch to one of the parallel algorithms when
Concurrency refers to the nature of the parallel dual simplex algorithm, not its execution. Hence, when using
I see that this is done in commercial solvers, so we'll discuss adding it. |
|
@jajhall thank you for the answers! I'm trying to understand what's the best approach in my situation (LP, control via CLI, parallel preferred but single-thread sometimes required for stability). My current plan is:
From your comment above, this “faux serial” approach can be slower than pure serial, but since we only use it by exception (as needed to avoid crashes only on some VMs), it shouldn't be a big problem. Does this sound correct? I would like to still have a CLI way to switch to serial (former |
Uh oh!
There was an error while loading. Please reload this page.
Version: v1.15.0
Call method:
highs.exeSolver: LP
OS: Windows 11
Hello, I'm trying to configure and toggle LP parallel mode via command-line, using guidance from this comment, and have the following observations:
simplex_strategy— no CLI parameter for this, only the settings file, right?When I set
simplex_strategy = 3(corresponding tokSimplexStrategyDualMulti) in the options file, the command line parameter--parallel onseems to be ignored whether it's present, absent, or "off". The timing is the same (parallel is faster than serial LP for me), and the log showsUsing parallel dual simplex solver - PAMI with concurrency of 8in both cases.My main question is how to properly toggle parallel/serial mode via the command-line now. If
--parallel off(or its omission) worked it would be clear, but for now I seem to be locked into parallel viasimplex_strategy=3, and no way to affect it from CLI.I need to toggle these programmatically, so editing the options file on the fly is my very last resort.
I also tested the new
--threads 1parameter — this is great, THANK YOU! The performance is affected (as expected), but, a minor feedback, the log may be misleading, as it gives a warning but then still states concurrency of 8 afterwards:PS May I suggest a minor improvement to
Highs.log— when adding to an already non-empty file, it would be useful to add a divider (e.g. a few empty rows) before each new run. And add a timestamp (date-time) around the run title (Running HiGHS…etc).All reactions