SMARTFIRE HARDWARE REQUIREMENTS

Quick Links

SMARTFIRE Hardware Requirements

Requirements for Parallel SMARTFIRE

SMARTFIRE Typical Run Time Performance

SMARTFIRE Parallel Performance

SMARTFIRE HARDWARE REQUIREMENTS

The SMARTFIRE software is available to run on PCs under Microsoft Windows 32 bit operating systems (e.g. Windows NT v4.0, Windows 2000, Windows XP, Windows Vista).

 SMARTFIRE V4.1 has not been tested or validated under all variants of Windows Vista and has not been validated on Windows Server 2003.

The minimum recommended PC hardware requirements are:

Additionally you will require a two or three button mouse in order to operate the various user interfaces.

Please contact the SMARTFIRE developers for information about obtaining support for other hardware platforms or operating systems.

SMARTFIRE requires a suitable text editor program (correctly installed and registered to handle files of extension “.txt”) in order to edit data files (in plain text format).

^[*1] Some of the Smartfire Menus might be slightly larger than the display size of SVGA. All of the menus should still be useable at this resolution.

^[*2] Smartfire has been developed using "Small Fonts". There might be menus within Smartfire that do not display properly when using "Large Fonts" selected from the Display Settings panel of your operating system.

^[*3] Intel Extreme Edition processors are likely to be faster than the mainstream Intel Core 2 Duo processors due to the larger caches and/or clock speeds, but this is untested. Recent AMD Phenom processors are untested at this time.

REQUIREMENTS FOR PARALLEL SMARTFIRE

The only additional complexity of running Parallel SMARTFIRE is that the user is required to select the networked PC nodes that will be used to run the simulation in parallel mode. It is up to the user to ensure that the PC nodes that are to be used for parallel SMARTFIRE have:-

• the required access permissions (administrator rights) for the user to run tasks,

• the appropriate MPI communications and tasking software installed on all of the PCs that are to be used (Note that not all of the PCs on the network need to have the MPI software installed – only those that will be used!),

• a suitably fast network between the machines to make parallel processing practical (typically 100Mbit/s, and preferably 1 Gbit/s with low latencies, parallel processing is possible for 10 Mbit/s networks but only for small parallel configurations (< 4 machines)),

• limited possibility of being switched-off, re-started or disconnected from the network for the duration of any simulations,

• only light local processing usage that will not interfere with (or be interfered with by) the parallel SMARTFIRE task running in the background,

• only the ‘Master’ processor (the computer that launches parallel SMARTFIRE) requires a dongle.

SMARTFIRE TYPICAL RUN-TIME PERFORMANCE  Back

The speed at which SMARTFIRE can perform simulations is dependent on the nature of the simulation and the computer hardware used. The primary factors, which influence this performance, are: the size of the computational mesh, the nature of the physics options selected, the computational options selected, the computer processor type (and speed) and the amount of physical RAM available to the software.

As a benchmark measure of performance the following scenario has been used:

Geometry:
Steckler room fire case, dimensions length=2.8m by width=2.8m by height=2.18m, open door measuring width=0.74m by height=1.83m centrally located on one wall. The room was meshed with two different cell budgets as (a) 12348 control volumes (cells) with directional cell budgets of nx=21, ny=21 and nz=28, and (b) 203742 control volumes with directional budgets of nx=66, ny=63 and nz=49. The centrally located fire is represented as a cube of edge 0.3m which has a steady heat release output of 59.9 kW or an equivalent fuel mass release rate of 0.001198 kg/s of Methane fuel (when using the gaseous combustion model).

Physics:
Transient turbulent flow with heat transfer and the compressible ideal gas law, using two radiation options i.e. six-flux (enhanced) radiation model and a multiple ray radiation model with 24 rays. In certain runs the heat release fire was replaced with a gaseous fuel release fire. Sixty time steps of duration 1 second (representing a total of one minute of simulated time) were simulated with a maximum limit of 50 sweeps per time step.

Solver options:
Domain SOR using 200 internal iterations for Pressure and 50 internal iterations for other scalar variables except for Momentum and Radiation which use only 5 internal iterations.

Run times:
This scenario has been run - using the above geometry, physics and solver options - on the Windows XP operating system with the following timings.

 Table: Typical hardware performance.

* Old timings using SMARTFIRE v3.0. Other timings using SMARTFIRE v4.0. It should be noted that SMARTFIRE v4.0 introduced significantly greater freedom to the inner solver iterations which produces better convergence at the expense of runtime performance. This is offset by the fact that SMARTFIRE v4.0 had a number of internal optimizations which speed up the processing.

An additional time study has been performed that is intended to represent typical simulations and cell budgets that may be encountered in practical engineering design applications. The scenario involves a three-storey building. in which the ground floor is sealed (no external ventilation) with stairway openings to the upper two floors and a large fire located in one of the corner rooms of the ground floor. The fire is represented using the combustion model with a heat release rate equivalent to 4 MW. The scenario has combustion, smoke and six-flux radiation models enabled. A refined mesh of 604,350 cells was used for the simulation. Note that this type of mesh would only be recommended for production runs, not for explorative analysis. The scenario was run for a total of three minutes of simulated time (using 90 time steps of 2 seconds per time step) and a maximum of 50 iterations per time step. The total run time, using a Pentium 4, 3.2 GHz (FSB 800 MHz) Windows XP PC, was 62 hours.

Finally, as an indication of the size of scenario that can be run on more limited PCs, geometries using 204,000 cells have been run on a Pentium III 733 MHz PC using windows NT 4.0 with 768 Mb (¾ GB) of physical memory using physics capabilities of flow, turbulence, heat and six flux radiation. The core memory used by the SMARTFIRE was reported as 389 MB. It should be noted that if the total memory used (by all applications, the operating system and SMARTFIRE running a simulation case) exceeds the available physical memory then the operating system will start paging to disk and the run time performance will slow to a crawl making the simulation practically unusable. A limitation of memory addressing range for Windows 32 Bit systems means that the maximum amount of physical memory that can be used by any application is 2 GB. This corresponds to approximately 780,000 cells in the SMARTFIRE software (depending on the physics options that are activated).

SMARTFIRE PARALLEL PERFORMANCE

Figure 1 - Parallel Speed-up for a network of Pentium III 800MHz PCs

Figure 2 - Parallel Speed-up for a network of Pentium 4 3.2GHz PCs

The speedup in run-time using Parallel SMARTFIRE depends on the size of the problem being run. As the overall problem size increases, so the communications overhead becomes less significant - when compared to the portion of the whole job being handled by each processor. Figure 1 shows the parallel speed-up for two different problem sizes (26,000 cells and 100,000 cells) on a network of Pentium III 800MHz PCs attached via a 100Mb/s LAN. Figure 2 shows the parallel speed-up for two different problem sizes (100,000 cells and 1,000,000 cells) on a network of Pentium IV 3.2GHz PCs attached via a 1Gb/s LAN. Even though the parallel speed-up will never reach the 1:1 ratio of the ideal curve, the whole job is still being processed in a fraction of the time that would be required for a serial simulation AND it is possible that a serial simulation is not possible due to the memory requirements of large cell budgets.

Figure 3 - The wall clock time required to run the 1 million cells case with a varying number of Pentium 4 PCs operating in parallel.

The 1 million cell problem was based on a test case for the nuclear industry. The domain had dimensions of 100m x 20m x 50m. The problem was run for 3 minutes of simulated time using 90 times steps of 2 seconds duration. The problem size of 1 million cells can not be accommodated on a single processor and the time taken on a single processor is estimated to be twice the time taken on two processors. This highlights a further advantage of parallel processing allowing the use of cell budgets beyond the capability of a single PC. 

It can be seen, from the Figure 3, that the problem would have taken 105 hours to run on a single PC and this is reduced to less than 15 hours when 8 PCs are utilised. This allows a Fire Safety Engineer to completely run the problem overnight instead of having to wait well over 4 days to obtain the solution.

Back to SMARTFIRE