Extremely long delay at the solver start from JuMP

[jspitz-git]

There is a long delay after calling JuMP.optimize! for any larger problem.

The code:

using JuMP, HiGHS, MacroEnergyScaling
m = read_from_file("medium.mps", format = MOI.FileFormats.FORMAT_MPS)
set_optimizer(m, HiGHS.Optimizer)
JuMP.set_optimizer_attribute(m, "solver", "simplex")
optimize!(m)

The output:

Running HiGHS 1.14.0 (git hash: 7df0786de3): Copyright (c) 2026 under Apache 2.0 license terms
Using BLAS: blastrampoline-5
......
 **5 minutes delay here**
.....
LP has 586972 rows; 420526 cols; 1446423 nonzeros
Assessing costs and bounds after applying user_objective_scale option value of -7
Coefficient ranges:
  Matrix  [1e-03, 1e+02]
  Cost    [8e-03, 8e+03]
  Bound   [0e+00, 0e+00]
  RHS     [1e+00, 6e+05]
Presolving model
238410 rows, 211200 cols, 696758 nonzeros 0s
202266 rows, 175079 cols, 583571 nonzeros 0s
197915 rows, 118168 cols, 568001 nonzeros 1s
Dependent equations search running on 37846 equations with time limit of 1000.00s
Dependent equations search removed 0 rows and 0 nonzeros in 0.01s (limit = 1000.00s)
197370 rows, 117623 cols, 571230 nonzeros 1s
Presolve reductions: rows 197370(-389602); columns 117623(-302903); nonzeros 571230(-875193) 
Solving the presolved LP
WARNING: Number of threads available = 6 < 8 = Simplex concurrency to be used: Parallel performance may be less than anticipated
Using parallel dual simplex solver - PAMI with concurrency of 8
  Iteration        Objective     Infeasibilities num(sum)
          0     0.0000000000e+00 Pr: 17520(2.83929e+08) 1.9s
      27853     4.9363762594e+04 Pr: 35833(2.26624e+10); Du: 0(3.32807e-06) 7.1s

I am using HiGHS.jl v1.23.0. The mentioned step lasts only a few seconds with HiGHS.jl 1.22.2. It happens with any larger problen both read from MPS file and built directly by JuMP. It does not happen with another solvers used for the same problem.

My system:

Sytem SKU LENOVO_MT_21H3_BU_Think_FM_ThinkPad L15 Gen 4
Processor	13th Gen Intel(R) Core(TM) i7-1355U, 1700 Mhz, cores: 10, logical processors: 12

Windows 11 PRO Czech version
Version	10.0.26200 Build 26200

Julia installation:

Julia 1.12.6

(@v1.12) pkg> status
  [87dc4568] HiGHS v1.23.0
  [4076af6c] JuMP v1.30.0
  [295af30f] Revise v3.14.2

Thanks.

Jiri

[odow]

Can you provide the MPS?

[odow]

Why does it say Assessing costs and bounds after applying user_objective_scale option value of -7?

Is that log output from your JuMP code? It seems like something else is going on.

[jspitz-git]

medium.zip

I forgot to delete the scaling option in the previous run, sorry for that. Now I tried it on another computer with a clean installation with the same result.

Script:
using Pkg
Pkg.activate("c:/tmp")
cd("c:/tmp")
using JuMP, HiGHS
m = read_from_file("medium.mps", format=MOI.FileFormats.FORMAT_MPS)
set_optimizer(m, HiGHS.Optimizer)
JuMP.set_optimizer_attribute(m, "solver", "simplex")
optimize!(m)

Output:
Activating project at c:\tmp
Running HiGHS 1.14.0 (git hash: 7df0786de3): Copyright (c) 2026 under Apache 2.0 license terms
Using BLAS: blastrampoline-5
LP has 586972 rows; 420526 cols; 1446423 nonzeros
Coefficient ranges:
Matrix [1e-05, 1e+02]
Cost [1e+00, 1e+06]
Bound [0e+00, 0e+00]
RHS [1e+00, 6e+05]
Presolving model
238410 rows, 211200 cols, 696758 nonzeros 0s
202266 rows, 175079 cols, 583571 nonzeros 1s
197915 rows, 118168 cols, 568001 nonzeros 1s
Dependent equations search running on 37846 equations with time limit of 1000.00s
Dependent equations search removed 0 rows and 0 nonzeros in 0.00s (limit = 1000.00s)
197370 rows, 117623 cols, 571230 nonzeros 1s
Presolve reductions: rows 197370(-389602); columns 117623(-302903); nonzeros 571230(-875193)
Solving the presolved LP
Using dual simplex solver
Iteration Objective Infeasibilities num(sum)
0 0.0000000000e+00 Pr: 17520(9.08572e+09) 2.3s
18677 4.1918333951e+06 Pr: 34982(1.33365e+12); Du: 0(4.36757e-05) 7.4s
22774 5.2829621215e+06 Pr: 36132(2.88785e+11); Du: 0(4.0564e-05) 12.6s

Status:
(tmp) pkg> status
Status C:\tmp\Project.toml
[87dc4568] HiGHS v1.23.0
[4076af6c] JuMP v1.30.0

[odow]

I can't reproduce. Can you show the output of this in a fresh session?

% julia --project=highs-337
   _       _ _(_)_     |  Documentation: https://docs.julialang.org
  (_)     | (_) (_)    |
   _ _   _| |_  __ _   |  Type "?" for help, "]?" for Pkg help.
  | | | | | | |/ _` |  |
  | | |_| | | | (_| |  |  Version 1.12.5 (2026-02-09)
 _/ |\__'_|_|_|\__'_|  |  Official https://julialang.org release
|__/    

julia> using JuMP, HiGHS

julia> model = read_from_file("/Users/odow/Downloads/medium.mps")
MEDIUM
├ solver: none
├ objective_sense: MIN_SENSE
│ └ objective_function_type: AffExpr
├ num_variables: 420526
├ num_constraints: 902358
│ ├ AffExpr in MOI.EqualTo{Float64}: 271589
│ ├ AffExpr in MOI.GreaterThan{Float64}: 10
│ ├ AffExpr in MOI.LessThan{Float64}: 315373
│ ├ VariableRef in MOI.GreaterThan{Float64}: 219026
│ └ VariableRef in MOI.LessThan{Float64}: 96360
└ Names registered in the model: none

julia> set_optimizer(model, HiGHS.Optimizer)

julia> set_attribute(model, "solver", "simplex")

julia> set_attribute(model, "time_limit", 0.01)

julia> @time optimize!(model)
Running HiGHS 1.14.0 (git hash: 7df0786de3): Copyright (c) 2026 under Apache 2.0 license terms
Using BLAS: blastrampoline 
LP has 586972 rows; 420526 cols; 1446423 nonzeros
Coefficient ranges:
  Matrix  [1e-05, 1e+02]
  Cost    [1e+00, 1e+06]
  Bound   [0e+00, 0e+00]
  RHS     [1e+00, 6e+05]
Presolve reductions: rows 586972(-0); columns 420526(-0); nonzeros 1446423(-0) - Timeout
Model status        : Time limit reached
Objective value     :  0.0000000000e+00
HiGHS run time      :          0.11
 20.046197 seconds (18.58 M allocations: 1019.245 MiB, 2.03% gc time, 0.41% compilation time: 20% of which was recompilation)

julia> set_optimizer(model, HiGHS.Optimizer)

julia> set_attribute(model, "solver", "simplex")

julia> set_attribute(model, "time_limit", 0.01)

julia> @time optimize!(model)
Running HiGHS 1.14.0 (git hash: 7df0786de3): Copyright (c) 2026 under Apache 2.0 license terms
Using BLAS: blastrampoline 
LP has 586972 rows; 420526 cols; 1446423 nonzeros
Coefficient ranges:
  Matrix  [1e-05, 1e+02]
  Cost    [1e+00, 1e+06]
  Bound   [0e+00, 0e+00]
  RHS     [1e+00, 6e+05]
Presolve reductions: rows 586972(-0); columns 420526(-0); nonzeros 1446423(-0) - Timeout
Model status        : Time limit reached
Objective value     :  0.0000000000e+00
HiGHS run time      :          0.11
 19.763519 seconds (18.29 M allocations: 1004.765 MiB, 3.11% gc time)

[odow]

Don't worry. 19 seconds is slow.

Here's v1.13.1

julia> using JuMP, HiGHS

julia> model = read_from_file("/Users/odow/Downloads/medium.mps")
MEDIUM
├ solver: none
├ objective_sense: MIN_SENSE
│ └ objective_function_type: AffExpr
├ num_variables: 420526
├ num_constraints: 902358
│ ├ AffExpr in MOI.EqualTo{Float64}: 271589
│ ├ AffExpr in MOI.GreaterThan{Float64}: 10
│ ├ AffExpr in MOI.LessThan{Float64}: 315373
│ ├ VariableRef in MOI.GreaterThan{Float64}: 219026
│ └ VariableRef in MOI.LessThan{Float64}: 96360
└ Names registered in the model: none

julia> set_optimizer(model, HiGHS.Optimizer)

julia> set_attribute(model, "solver", "simplex")

julia> set_attribute(model, "time_limit", 0.01)

julia> @time optimize!(model)
Running HiGHS 1.13.1 (git hash: 1d267d97c): Copyright (c) 2026 under Apache 2.0 license terms
Using BLAS: blastrampoline 
LP has 586972 rows; 420526 cols; 1446423 nonzeros
Coefficient ranges:
  Matrix  [1e-05, 1e+02]
  Cost    [1e+00, 1e+06]
  Bound   [0e+00, 0e+00]
  RHS     [1e+00, 6e+05]
Presolve reductions: rows 586972(-0); columns 420526(-0); nonzeros 1446423(-0) - Timeout
Model status        : Time limit reached
Objective value     :  0.0000000000e+00
HiGHS run time      :          0.11
  1.651113 seconds (18.58 M allocations: 1019.240 MiB, 21.29% gc time, 5.26% compilation time: 24% of which was recompilation)

(hgh2) pkg> st
Status `/private/tmp/hgh2/Project.toml`
⌃ [87dc4568] HiGHS v1.22.2
  [4076af6c] JuMP v1.30.0
  [c46f51b8] ProfileView v1.10.3
Info Packages marked with ⌃ have new versions available and may be upgradable.

[odow]

I've reported this upstream: ERGO-Code/HiGHS#2977

I'm going to close this issue because there isn't anything we can do in HiGHS.jl about it. For now, just stick with HiGHS@1.22

[jspitz-git]

Here are reults on AMD Ryzen 7 5700G with Radeon Graphics, 3801 Mhz, cores: 8, logical processes: 16:PS

C:\tmp> julia --project=/tmp
               _
   _       _ _(_)_     |  Documentation: https://docs.julialang.org
  (_)     | (_) (_)    |
   _ _   _| |_  __ _   |  Type "?" for help, "]?" for Pkg help.
  | | | | | | |/ _` |  |
  | | |_| | | | (_| |  |  Version 1.12.6 (2026-04-09)
 _/ |\__'_|_|_|\__'_|  |  Official https://julialang.org release
|__/                   |

julia> using JuMP, HiGHS

julia> model = read_from_file("medium.mps")
MEDIUM
├ solver: none
├ objective_sense: MIN_SENSE
│ └ objective_function_type: AffExpr
├ num_variables: 420526
├ num_constraints: 902358
│ ├ AffExpr in MOI.EqualTo{Float64}: 271589
│ ├ AffExpr in MOI.GreaterThan{Float64}: 10
│ ├ AffExpr in MOI.LessThan{Float64}: 315373
│ ├ VariableRef in MOI.GreaterThan{Float64}: 219026
│ └ VariableRef in MOI.LessThan{Float64}: 96360
└ Names registered in the model: none

julia> set_optimizer(model, HiGHS.Optimizer)

julia> set_optimizer_attribute(model, "solver", "simplex")

julia> set_attribute(model, "time_limit", 0.01)

julia> @time optimize!(model)
Running HiGHS 1.14.0 (git hash: 7df0786de3): Copyright (c) 2026 under Apache 2.0 license terms
Using BLAS: blastrampoline-5
LP has 586972 rows; 420526 cols; 1446423 nonzeros
Coefficient ranges:
  Matrix  [1e-05, 1e+02]
  Cost    [1e+00, 1e+06]
  Bound   [0e+00, 0e+00]
  RHS     [1e+00, 6e+05]
Presolve reductions: rows 0(-586972); columns 0(-420526); nonzeros 0(-1446423) - Timeout
Model status        : Time limit reached
Objective value     :  0.0000000000e+00
HiGHS run time      :          0.02
182.886079 seconds (18.60 M allocations: 1017.143 MiB, 0.72% gc time, 0.10% compilation time: 20% of which was recompilation)

julia> set_optimizer(model, HiGHS.Optimizer)

julia> set_optimizer_attribute(model, "solver", "simplex")

julia> set_attribute(model, "time_limit", 0.01)

julia> @time optimize!(model)
Running HiGHS 1.14.0 (git hash: 7df0786de3): Copyright (c) 2026 under Apache 2.0 license terms
Using BLAS: blastrampoline-5
LP has 586972 rows; 420526 cols; 1446423 nonzeros
Coefficient ranges:
  Matrix  [1e-05, 1e+02]
  Cost    [1e+00, 1e+06]
  Bound   [0e+00, 0e+00]
  RHS     [1e+00, 6e+05]
Presolve reductions: rows 0(-586972); columns 0(-420526); nonzeros 0(-1446423) - Timeout
Model status        : Time limit reached
Objective value     :  0.0000000000e+00
HiGHS run time      :          0.02
178.100165 seconds (18.29 M allocations: 1002.119 MiB, 0.76% gc time)

Julia>

[jspitz-git]

I've reported this upstream: ERGO-Code/HiGHS#2977

I'm going to close this issue because there isn't anything we can do in HiGHS.jl about it. For now, just stick with HiGHS@1.22

Thank you.

[odow]

No problem. Thanks for reporting. These things are quite hard to notice otherwise.