I have 4GB of RAM using Slackware 12.2, with a recompiled kernel.

I have read there is a performance hit when configuring the kernel to use PAE to recognize more than 4GB of RAM. How much of a performance hit I wonder?

I would appreciate links to discussions providing actual data. I find only the claim but no data.

Thanks.

I took this from a red hat white paper, its the closest thing I could find. Here is the link to the whole paper http://people.redhat.com/nmurray/RHE...whitepaper.pdf


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%.

When I changed my kernel some months ago to use the 64G flag I tried to find actual data as well.
I understand perfectly that the memory switching will cause some overhead, but personally I have not 'felt' any differences between the two kernels.

I started a search for some reliable benchmarking tools but gave up after a while.

Benchmarking a system is not trivial:
- do you want to test memory access?
- disk I/O?
- network I/O?
It all depends on what you use your server for.

I have seen servers with 20 - 30% processor usage but terrible response times because of long disk I/O queues or network latency problems.
I have seen processors cooking at 100% usage because of badly written database apps.

And I know of dozens of 32-bit servers (Windows & Linux) using PAE and performing very well and other 64-bit servers with lots of memory only causing performance head-aches.

By the way, managing data centers is my job

Thanks for the info.

I'm just a basic desktop user. A 1% performance hit is insignificant for my usage. I suppose I could temporarily use a kernel without PAE and then test for a few days (gut feel --- nothing methodical or scientific), but at an estimated 1% hit for most usage, I doubt I'd see any difference.

no discernible difference here

just switched to a PAE kernel last week though

I re-compiled my kernel a few days ago to PAE also. I have not seen any effects on performance so far. I am also a simple desktop user.

I believe I saw that it is only slower to initialize.

I would doubt that. Having to move any data above 1 Gig down to kernel addressable storage would be the issue.
Who does heavy I/O with big writes ... ?

Hi,

Slackware 12.2 2.6.28.8 8 GB ram Intel dual core with virt. support

I've been running the huge mem support for near 8 weeks or longer now.

I was totally unaware that it could have a performance hit until I saw this thread.

Windows 7 beta runs very very fast in KVM with near 1850 megs of ram specified to it. Ditto Win XP SP3 with a gig of ram to it. Both of these run too fast (no complaint, works for me).

Slackware runs fast no matter whether KVM is in use or not.

In slack, Firefox 3 tabs sometimes hesitate before they close (after I've clicked the X on the tab to close it). Not a bother, but weird to me, I don't know what causes. Maybe it my ATI catalyst 9.3 though my video mostly works well.

--
Alan.

I've never compared them using proper benchmarks. But the system feels about the same for me, using 2.6.28 and PAE.