Trying to use UDF .xll to parallelize Excel on HPC grid

Question

I'm looking to parallelize a Excel calculation on a HPC grid, I'm using Excel 2007 so I'm trying to multithread it through the UDF in a XLL because it is the only way (at least for 2007).

I would like to code the XLL in C# and as this thread suggests I can use exceldna.

Yet I can't understand how should I multithread it. Do I have to re-write all the VBA function in C# while multithreading it ? Can I just create several threads, each one of them calling one VBA macro ?

Answer 1

Excel-DNA allows different approaches that you might find useful:

• you can make UDFs marked as 'thread-safe' to use Excel's support for multithreaded calculation - in this case Excel controls the threading, and will call your UDFs concurrently from different threads,
• you can use the asynchronous functions in Excel-DNA (possibly using the .NET 4 Task API) to start the work on the grid, and upon completion notify Excel to recalculate,
• you can use the Reactive Extensions support in Excel-DNA and build a calculation pipeline on the Rx library ,
• you can use the support for Windows HPC clusters built into Excel 2010, using Excel-DNA to host the managed code in the client and on the cluster nodes.

In none of these cases would it make sense to call into Excel VBA - Excel will always ensure that the VBA code is running on Excel's main thread.

If your function code is in VBA, you might consider using VB.NET for your Excel-DNA add-in, instead of C# - it might make your move to .NET quite easy. You could also make a mixture of C# and VBA in your managed Excel-DNA add-in.