Performance using paging based non-executable pages

[l0ud]

Hello all,

After reading the documentation I know that using the paging based non-exec feature will result in some performance degredation, but could someome please inform me if what I am experiencing is normal?

After enabling the paging based I noticed the machine seemed to run excessively slower; so, I ran some tests compiling the 2.4.23 kernel.

Compiling the 2.4.23 source from a .23 kernel with Segmenation:
real 2m14.555s
user 7m24.220s
sys 0m21.310s

Compiling the 2.4.23 source from a .23 kernel with Paging:
real 62m33.588s
user 50m20.560s
sys 193m8.860s

I would say that is way more than a "slight" impact! Is this on par with what others have experienced? The machine is a dual Xeon 2.4Ghz with HyperThreading enabled. Its running on an Intel SE7501CW2 board with 2GB of ram.

Thanks!

[siti]

Wow that is a huge difference!

It probably made your dual xeons have the same performance as dual pentium 3's

[PaX Team]

I would say that is way more than a "slight" impact! Is this on par with what others have experienced? The machine is a dual Xeon 2.4Ghz with HyperThreading enabled. Its running on an Intel SE7501CW2 board with 2GB of ram.

i had the first report of this P4 behaviour more than a year ago but never debugged it to figure out what it doesn't like in PAGEEXEC, especially as i was about to finish SEGMEXEC which obsolated PAGEEXEC on i386 for most uses. if you feel like doing some experiments let me know in email (no guarantees that it can be fixed, the entire TLB stuff is very much undocumented).