[immortality]

cache
{
    grad(U);
}

[wyldckat]

Hi Ehsan,

A bit more context to the question would make it a lot easier to help... in other words, how is this being used?
In even other words, how can I test what this does, if you don't indicate how you've tried to use it!?

Best regards,
Bruno

[immortality]

Hi
I saw it in a case I was curious about only that what does.never mind.

[wyldckat]

Then I guess I'll have to give a generic answer ... the term "cache" is explained in Wikipedia: http://en.wikipedia.org/wiki/Cache_%28computing%29

In computer science, a cache (/ˈkæʃ/ KASH)[1] is a component that transparently stores data so that future requests for that data can be served faster. The data that is stored within a cache might be values that have been computed earlier or duplicates of original values that are stored elsewhere. If requested data is contained in the cache (cache hit), this request can be served by simply reading the cache, which is comparatively faster. Otherwise (cache miss), the data has to be recomputed or fetched from its original storage location, which is comparatively slower. Hence, the greater the number of requests that can be served from the cache, the faster the overall system performance becomes.

[Tobi]

Hi all,


I am just interested which fields should be cached and when? It is obvious that recalculating a field requires much longer than caching it after the first evaluation. However, how can we determine, which field we should cache and which not?

• If I would build my own solver e.q. a stress solver for linear elastic problems, the grad(U) field has to be evaluated several times. Here, I guess the cache option for grad(U) would be speed up the calculation
• While using FOAM standard solvers, how can we determine which field to cache and which not?

Is there any rule or any hint? E.g., we could cache actually each field
Okay, if it is just needed once, then it will probably decrease the calculation speed a bit based on caching the data always. However, it was just an example.

[jwstolk]

Originally Posted by Tobi

It is obvious that recalculating a field requires much longer than caching it after the first evaluation.

Only when using memory takes no time. With recent CPU's the calculation times and the number of cores has improved much faster than memory bandwidth. CPU-cache is practically irrelevant since OpenFOAM cycles through a large dataset that is usually larger than could fit in the CPU-cache.


If the memory bandwidth is the bottleneck on your hardware, it may be faster to re-calculate the value instead of saving a copy. It also depends on how difficult it is to re-calculate the value. what works for "U" may not work for other values.


Take a run that is typical of what you are using the hardware for (cell size, solver selection, decomposition, etc.) and run that twice, with only the cache U disable as difference between the settings, and check which one runs faster.


Same for hyper-threading: hyper-threading usually makes CFD's run slower but I had some quad-core systems with quad-channel RAM. On that specific hardware, hyper-threading enabled (8 threads) was slightly faster. Just test it with a typical workload.

[ufocfd]

good question - do you need to cache the grad(U) in simpleFoam?

Originally Posted by immortality

cache
{
    grad(U);
}

[Tobi]

No you don't. You don't need to cache any quantity but if it is used several times, its much cheaper to save the first calculation and simply take the results in any other place.