Arch kernel with PAE

Question

I asked some times ago about including PAE (HIGHMEM) in the offical Arch Kernel in x86 and i was said no beceause it is slower and when i want more than 4gb memory i should use x86_64
now i seen. 32bit is not slower with PAE. Only the stuff over 3GB is slower than with 64bit.
So the argoument PAE is slower is nonsense beceause the memory which is accessed slower is not existing without PAE so we have the choose between
No Memory over 3GB
Slow Memory over 3GB or lemme say not fast as it can be
the first 3GB are the same speed with and without PAE so maybe we should re-think about this
Last edited by Vamp898 (2009-10-20 00:32:29)

Answer 1

Allan wrote:
That is very strange (and dare I say wrong) given how PAE works...   it is not as if it uses it memory mapping only when needing greater than 3Gb of memory, it needs to use it all the time:  http://en.wikipedia.org/wiki/Physical_Address_Extension
And here is the only real attempt at benchmarks I have seen.  Although it is a bit old...
http://people.redhat.com/nmurray/RHEL-2.1-VM-whitepaper.pdf wrote:The performance impact is highly workload dependent, but on a fairly typical kernel
compile, the PAE penalty works out to be around a 1% performance hit on Red
Hat's test boxes. Testing with various other workload mixes has given performance
hits ranging from 0% to 10%.
wow its really a bit old and i dont know how it was tested
is there a really good way to benchmark it?
i just written a bash skript which pumped up the memory and taked the time until it needed to reach 3,8GB
Last edited by Vamp898 (2009-10-20 02:30:17)