I was under the impression that only 2GB was mapped to userspace, and the
other 2GB was mapped for kernel data, although I could be wrong.
I'm fairly sure redhat's kernel configures a 3G/1G split. So the max you could ever get would be 3G, also it will depend on which version of glibc you are using. You can push this to 3.5 by hacking the kernel a bit. Take a look at Andrea's kernel:
ftp://ftp.kernel.org/pub/linux/kernel/people/andrea/kernels/v2.4/2.4.22pre7aa1/00_3.5G-address-space-5
Jon
On Tue, 5 Aug 2003, Robert Vaughn wrote:
During some tests we have observed some odd memory behavior in Linux. It appears that our Linux server with 10 GB of RAM will only allocate a maximum of 2.8 GB per process. When we try to exceed 2.8 GB per process the process dies. We are interested in finding why and how to fix the behavior.
Our questions are... 1) What is limiting the amount of memory that we can access? I understand that we should not be able to access above 4GB per process on a 32 bit system. However, being able to access only 2.8GB is not very good. 2) What can we do about the rather low memory limit?
Following is some information about what we have been doing. I can provide additional details as requested.
The program... The program that we're using to perform the tests is a Perl script that consumes a specified amount of memory through a loop of stuffing characters into an array. I wrote the Perl script. We are going to create a C++ version of the Perl script. However, we do not think that Perl is the problem. Previously I have used the same script to consume about 3.8 GB of memory before the script gets killed. When the script runs we observe no swap behavior. We can run multiple scripts of say 2GB memory consumption and eat up the entire free memory space and then start eating into swap.
Background... Intel Xeon 8X 10 GB ram Redhat As 2.1, 2.4.9-e.3enterprise
Memory Info with No Load on Server... Total: 10303272 KB Used: 315512 KB Free: 9987760 KB
-- Once you have their hardware. Never give it back. (The First Rule of Hardware Acquisition) Sam Flory <[EMAIL PROTECTED]>
-- redhat-list mailing list unsubscribe mailto:[EMAIL PROTECTED] https://www.redhat.com/mailman/listinfo/redhat-list