On 20 Mar 2000 16:24:39 +0100, you wrote:
>Den 20-Mar-00 13:39:26 skrev James Sutherland følgende om "Re: Overcommitable memory??":
>> On 19 Mar 2000 19:27:22 +0100, you wrote:
>
>>> Do you actually have any numbers to back up your claims?
>
>> Do you have any justification whatsoever for your opposition to demand
>> allocation of memory?
>
> OK, so you have no numbers to back up your claims. Would you please at
>least try to get some? Otherwise there is little point in discussing your
>doomesday swap predictions.
I do have (rough) numbers from a large WWW site I ran quite recently.
I often had 100 or more processes running at once, alongside many
other jobs (including satellite image processing) on a box with just
over 1Gb of VM. Without overcommit, the processes would not have
fitted in VM - even if I had killed every other process running.
> As for the justification of not overcommitting memory, can I assume that
>you have heard of the ANSI/ISO C and POSIX standards?
Yes, I have. Why do you want to remove one of the key benefits of Unix
platforms?
>>> A programmer can waste quite a lot of time hunting for non-existant
>>>bugs to explain crashes caused by a kernel which can't keep count of its
>>>memory. Been there, done that.
>
>> If your app takes all the available memory, it should be blatantly
>> obvious that you have done something wrong.
>
> Not in any way. The application can't read the kernel's mind. The system
>has x MB of RAM+swap, the application asks for x+20 MB, the kernel says
>"OK". How is the application supposed to know that the kernel is telling a
>blatant lie? If you can't trust the kernel, your system is fsck'ed.
The kernel doesn't do that anyway (unless you have explicitly disabled
sanity checking). With or without normal overcommit, the behaviour
here is identical; the only difference is in whether or not you really
populate the whole block of address space you requested BEFORE you use
it.
James.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Thu Mar 23 2000 - 21:00:30 EST