When to Use Parallel Simulations — What’s Different?
Simulations can now be faster by using the new parallel simulation mode in ApacheSim. This mode divides the simulation into either the number of cores that are available or the user defined number of months that a user defines.
The number of months are: 1, 2, 3, 4, 5, 6. If a 12 month annual simulation was divided up using the 1 month option there would be 12 individual months ran. The 2 month option would run 6 sets of two months.
The feature then divides that simulation up into the number of required runs, regardless of the simulation period chosen. For each simulation there is a pre-conditioning period set, so there would never be a x12 speed up. Reducing the pre-conditioning period can further increase the speed up, but that could compromise accuracy – particularly in heavier-weight buildings.
Found in ApacheSim simulation dialog
Simulation options (then)
Global preferences (then)
Parallel simulation mode.
When all of the simulations are complete the ‘divided’ simulations are joined back together to the original desired size of the simulation period.
A user may choose to view a partial set of results and divide a simulation up (even if it strictly isn’t the quickest) as the divided partial results files can be downloaded into VistaPro for analysis whilst the other results files are running – this would allow the user to figure out if there were any issues with the simulation set up, if there was the rest of the simulations could be stopped, if not the simulations can continue.
Benefits
Speed
Flexibility of approach
Ability to analysis / check partial result files
Limitations
1 ApacheSim license per 4 cores – so unless a user has 3x licenses & they have a 12 core machine – they will not be able to get the best out of this enhancement.
Separating the simulation can lead to differences between a full 12 month simulation and one that has been run as 12 single months (or any variation) – users must only use this if they can be confident that they are happy with the understandable differences.
For a final compliance run for example it may be considered appropriate to not use this feature.
However during evaluation and analysis it may be considered appropriate to use the feature.
Items to consider
Changes in temperature ‘at the joins’
Non-accumulation of (say) CO2 ‘at the joins’
Users should consider these carefully before using on every simulation.
General understanding of the performance
Parallel is always faster than serial on 12 / 4 / 2 cores:
Max speed up is 85 / 70 / 48%
Average is 65 / 61 / 43%
Cores always equal Speed as models get larger
4 cores is optimum on v.small models (where processing time of 12 cores starts to take over)
v.small models with other complexities (Detailed HVAC) will see better speed ups – although there isn’t strong correlation here
Error (for total energy) is greater the more you split the run up
Error x10 for 12 cores on average versus that of 2 cores
Error on average still low for 10 day pre-con & 12 cores ~0.3%
Error (for total energy) directly related to pre-con period
1 day pre-con ~10x worse than 10 days
Smaller pre-conditioning periods always give bigger errors. Should be avoided as there is no correlation for small pre-con and speed – probably due to additional instability in Apache.
