Re: Problem with malloc on very large blocks of memory

[Dan Klebanov]

> What else do you expect? You also have error message in the kernel
> log saying that system run out of memory and killed biggest memory
>
>  

What I expect is that malloc should behave according to spec.  In other 
words, if there's not enough memory left in the system,  then it should 
return NULL.  Otherwise, what's the point of checking for NULL?  The way 
it works now is that it hands you the memory, but barfs as soon as you 
try to use it.  

The little test program that I wrote was just meant as an extreme way of 
proving my point.  I only became aware of this issue when our software 
was mysteriously crashing, and then it turned out that the cause was a 
broken swap partition on the offending machine. The program was trying 
to allocate a 40 meg chunk of memory and it would die during memset. 
(Granted, our software was written by monkeys typing at random, and 
there's all sorts of memory leaks to show for it.)

> hog to stay alive. Just do not do that - disable memory overcommit
> if your working set may be larger than available memory and you cannot
> accept killing process.
>                                                Petr Vandrovec
>                                                address@hidden
                                               

Isn't it  the operating system's job to tell me when I'm out of memory?