New defaults

[totto82]

I think it is time to change some defaults in Flow related to accuracy and performance.
Motivation:

• $10^{-6}$ mb tol sometimes leads to too large accumulated mass balance errors
• drift compensation sometimes leads to significant convergence issues (and are not needed if mb = $10^{-7}$)
• ilu0 leads to tool many iterations for mb $10^{-7}$
• for cases where we force small timesteps (for instance using TUNING) 1 newton iteration is often sufficient)

I have run with these parameters for many cases for a long time.

This will need extensive testing before it can be merged.

[totto82]

jenkins build this please

[GitPaean]

not sure about the best way to do this. Do you think it might be easier to get things in if you split them among multiple PRs? The combined effects can be tricky to test.

[totto82]

not sure about the best way to do this. Do you think it might be easier to get things in if you split them among multiple PRs? The combined effects can be tricky to test.

I see your point, but

1. I would like a common discussion about the new defaults
2. They are all interlinked (the fourth one could be taken independently)
3. They all effect results, and it will be a large effort to check all the failed tests.

[totto82]

benchmark please

[alfbr]

From our end, this is a highly welcome initiative. We will probably need some time to do proper testing as this will impact all models we run.

[totto82]

for cases where we force small timesteps (for instance using TUNING) 1 newton iteration is often sufficient)

The change to the minimum newton iterations created some strange failure. As suggested by @GitPaean I will split that out to another PR.

There are 6 significant test failures that needs to be addressed.

1. MICP. Seg-faults with CPR
2. 2D_POLYMER_INJECTIVITY also have problems with CPR
   3-4) SPE1CASE1_BRINE and SPE1CASE1_PRECSALT starts oscillating takes forever to complete with mb=1e-7.
   5-6) NETWORK-01-RESTART and NETWORK-01-REROUTE-RESTART related to restart.

The rest is either minor or needs more careful examination of the results.

[totto82]

The MICP case seg-faults in extractCPRPressureMatrix with CPRW (works for the other variants) The MICP model is a one-phase (water) model with 5 components.

[alfbr]

benchmark please

[totto82]

SPE1CASE1_BRINE and SPE1CASE1_PRECSALT starts oscillating takes forever to complete with mb=1e-7.

For this case the simplest solution is to introduce a relaxed_tol_mb (ala TUNING 2-7).

[blattms]

Posting missing ytelses results:

Test	Configuration	Relative
opm-git	OPM Benchmark: drogon - Threads: 1	0.998
opm-git	OPM Benchmark: drogon - Threads: 8	0.999
opm-git	OPM Benchmark: punqs3 - Threads: 1	0.982
opm-git	OPM Benchmark: punqs3 - Threads: 8	1.006
opm-git	OPM Benchmark: smeaheia - Threads: 1	0.963
opm-git	OPM Benchmark: smeaheia - Threads: 8	1
opm-git	OPM Benchmark: spe10_model_1 - Threads: 1	1.003
opm-git	OPM Benchmark: spe10_model_1 - Threads: 8	1.003
opm-git	OPM Benchmark: flow_mpi_extra - Threads: 1 - FOIT (Total Oil Injection At End Of Run)	1
opm-git	OPM Benchmark: flow_mpi_extra - Threads: 8 - FOIT (Total Oil Injection At End Of Run)	1
opm-git	OPM Benchmark: flow_mpi_norne - Threads: 1	0.995
opm-git	OPM Benchmark: flow_mpi_norne - Threads: 8	1.009
opm-git	OPM Benchmark: flow_mpi_norne_4c_msw - Threads: 1	1.006
opm-git	OPM Benchmark: flow_mpi_norne_4c_msw - Threads: 8	0.999

• Speed-up = Total time master / Total time pull request. Above 1.0 is an improvement. *

[totto82]

jenkins build this please

[totto82]

jenkins build this please

[akva2]

jenkins build this please