How to speed up an ECLIPSE run
Sometimes the simulator can only solve the model by taking extremely short timesteps or excessive amounts of cpu time. Making small changes to the model or adding new options or new keywords can sometimes lead to dramatic improvements in the speed of the simulation without changing the results to engineering accuracy.
If the data cannot be changed, convergence criteria can sometimes be chosen to improve simulator performance. There is usually no need to change convergence criteria, and it is advised not to do so unless necessary; internal parameters in ECLIPSE will usually give a stable and robust solution for the model, in a reasonable cpu time, although of course a big complex model will take longer to run than a small simple model. In some cases, however, the results of a run can depend on the convergence criteria and how they are applied, and there is a trade-off between accuracy and speed.
Recognizing and correcting the cause of convergence problems is an important part of simulation. Generally:
- Failure to converge is not (usually) due to a bug in ECLIPSE. It is usually related to the data.
- Failure to converge Linear iterations is not always something to worry about.
- Failure to converge a minimum timestep can affect the validity of the results.
ECLIPSE can output a lot of messages, warnings, etc. The reservoir simulation engineer need to know what to do about them, which ones can be safely ignored, and which ones need some action. One can also ask ECLIPSE to produce reports showing how both linear and non-linear iterations are proceeding and the methods by which time steps are selected.
Most of the advice and suggestions apply equally to all the simulators. Note, there are differences in the detailed treatment between ECLIPSE Black Oil and ECLIPSE Compositional.
In fact all runs can be speeded up. There are three general solutions:
1. Hardware solutions:
- Get a faster machine. A new computer may be twice as fast as the equivalent computer from two years ago.
- Make sure no other jobs are running on your computer. If you are running on a cluster, check that there is no contention for memory or other resources on the nodes on which your job is running.
- Run in parallel. This will not speed up your run if there are convergence issues in serial mode. If your model is running smoothly but slowly on one processor, try running in parallel on two. If it speeds up, try running on four processors.
2. Reduce the size of the model:
- Can you reduce the number of grid blocks without affecting the quality of the solution? For instance can you replace a large number of water-filled blocks with an analytic aquifer?
- If you are running in compositional mode, can you model your fluid with fewer pseudo-components?
3. Identify data issues:
- Check ECLIPSE Convergence manual
- Can you gain any speed by changing the time-stepping?
- Can you identify the cause of any convergence problems?
Additional: ECLIPSE 300 Black-Oil. Did you know that an ECLIPSE Black-oil license allows you to run ECLIPSE 300 in black-oil mode? ECLIPSE 300 sometimes gives you more information and insight into a problem than ECLIPSE 100.