Why not use the publicly available 1000genomes data (or subset thereof)?

http://www.1000genomes.org/data

They have sequence data from various platforms, including Illumina. You can work directly on the sequence data, or pick the pre-aligned BAM files to work with.

If you want to speed up throughput for pipeline development, just subset the data into a single chromosome and its associated mapped reads.

EDIT: OK didn't see the part about association testing. I'm not sure what the requirements would be for that in terms of sample numbers/size, but you might still be able to leverage the 1000Genomes datasets that exist.

Looking at papers in the area, I guess one approach would be to download or generate a few sets of reads and then artificially introduce a known causal mutation or two into the groups as appropriate. It is an over-simplification but still useful for test purposes I believe. Exactly how to introduce the mutation into the data sets is another question entirely I suppose!

If I use 1000genome data, can I simulate pooled sample in some way and generate in it SNP&indel ?