Timing large file reads

[redskull]

Plain HeapAlloc does not zero memory. However, if you request more than 126 pages, i.e. more than about 520,000 bytes, Windows switches to VirtualAlloc and zeroes memory.

The heap function do zero memory, in a way; all pages come back from the memory manager zeroed out, and then the heap itself fills it with magic numbers so it can test for corruption.  Even if you only HeapAlloc a couple bytes, it still has to VirtualAlloc an entire page at a time to your process, so there's nothing stopping you from reading or writing to locations your haven't officially allocated, and causing bugs  So really, HeapAllocs are twice as slow, because it not only has to be zeroed from the memory manager, but then filled with the check characters from the heap.  As to why the Heap doesn't do this for large allocations...?  It seems to the tipping point of when it runs out of space in the intial heap, and has to allocate a new one somewhere else in memory

-r

[jj2007]

You are right - HeapAlloc returns BAAD F00D...

By the way, we covered that ad nauseam in this thread; which showed (not surprisingly) that HeapAlloc was considerably faster than VirtualAlloc for repeated small allocations.