I noticed that calculating the md5 sums, for large files, take much less time the second time you run the md5 command on them. Is there any way to clear the cache? I'm actually writing a parallel application that calculates many md5 sums and would like accurate measurements without caching.
The speed is not from an md5 cache, but rather from the file itself being cached in unused memory. This is automatically kept up to date, and dumping the disk cache will only degrade your system performance.