On Sun, Feb 28, 2010 at 5:48 PM, Ingo Weinhold <ingo_weinhold@xxxxxx> wrote: > A simple attribute lookups shouldn't be *that* expensive either. It would > basically be one additional syscall per file and the concerned block would That mostly depends on whether such an attribute fits in the small data area or not, which is probably the case Axel is afraid of, since if it isn't, it would (as you know) require several extra disk seeks/reads to go find the attribute dir, look up the attribute and then go read the attr data. > very likely be already cached, since Tracker reads a lot more info for the > files anyway. Unless those thousands of file names actually need to be > translated, the additional overhead is probably not even measurable. > Doesn't harm to actually test this, though. Excepting the case of things like Audio:, etc. attributes, most of Tracker's stuff fits in small_data though so nothing more than the inode block itself ever gets read for those. Regards, Rene