Quote:
Originally Posted by
Corona688
Let me get this straight... It wastes cycles in advance, to waste swap in advance, so processes can allocate vastly more memory than they need and just not instantiate it, to avoid overcommit problems. In what sense is this conservative?
Overcommit may be dangerous in some circumstances, but there isn't enough disk space in the world to swap-back a 64-bit address space.
Hi Corona688,
It does not waste cycles, it just reserves the appropriate amount of paging space in one go when a process starts, so it does not have to do it in small bits for every page it needs in strained memory conditions when there may also be a lot of CPU usage. In this sense it is conservative. It does not actually use the paging space. it just reserves it. These are just design choices, each with their own pros and cons.
Also, this is not about inefficient apps, but about reaping the benefits of demand paging and efficient memory use, or in other words, not needlessly wasting RAM.
---------- Post updated at 10:34 PM ---------- Previous update was at 09:25 PM ----------
Quote:
Originally Posted by
jlliagre
HP-UX has indeed this bizarre approach by default and is the only current Unix lacking virtual swap. This design limitation can be partially overcome by enabling pseudo-swap.
Overcomitting memory is a different and dubious beast.
Hi jlliagre,
Actually pseudo-swap is the default on HP-UX systems. I think virtual swap or pseudo-swap is a terrible idea, especially if you make it the default. It leads admins into thinking that they do not need to allocate swap space, while there are penalties involved in doing so.
It effectively means you are using memory for paging space, well basically you are marking pages as non-pageable. This part of memory will then fill up much more quickly than usual, so in fact you are throwing away the benefit of demand-paging while keeping its downside.
Also what happens under load is that only a small part of memory actually gets paged out and if paging then occurs, then instead of a smooth transition it will be much more abruptly with the most recent processes taking the brunt.
In other words the system will waste copious amounts of memory and becomes less stable under load. The only time it is useful is when you have much more internal memory then you are ever going to need or you do not have room for swap space.
S.