PDA

View Full Version : Page Faults?



[DSLR]MntlCase
04-09-2002, 09:22 PM
Howard,

Why does the TEN client generate so many page faults?

That basically means the instruction that generated it has to "re-execute" correct?

If so.....WOW!

Brian the Fist
04-09-2002, 10:07 PM
Yes, I noticed that too. Although it sounds fairly bad, and despite my lack of knowledge about nt/2000/xp, it appears that a page fault, in this context, means Windows tried to retrieve a requested piece of data from virtual memory (your swap file) but failed, so it had to load it from disk instead.

It just means Windows is not making good use of its swap file :jester: as the software does indeed do a lot of disk reading (as opposed to using a huge amount of RAM which is the alternative). Putting everything in a RAM disk gives you the best of both worlds..

Chris Wolfe
04-10-2002, 04:13 PM
Close. My understanding is that it is when windows tries to read data from a page of RAM and the data is not there. The terminology is kinda confusing because windows creates pages in ram and also in virtual memory via pagefile.sys. Anyway, it seems that the frequency of page faults corresponds to the updating of filelist.txt. Might be room for improvement there. Just an observation, I'm no programmer.

ColinT
04-10-2002, 04:24 PM
I watched my Task Manager also. The program generates 25,000-100,000 page faults per second. It just keeps going:)

I am at 1 billion 200 million now, with a 48 hour uptime.

Bone stock P4 2000 with 512 ram and a Raid 1 with two 40 gig drives.

ColinT
04-10-2002, 04:38 PM
I might have something here. I have an XP1800 running 2K also. I checked it. It's been up for 26 days now. It has accumulated a quarter Million pagefaults for Foldtraj.

Could we be talking about pipeline misses? The P4 has an awful architecture.

Dogbreath
04-16-2002, 09:32 AM
I don't think so, a pipeline issue would be dealt with purely in hardware, it's not the kind of thing an OS would know about AFAIK.

My 1.4 Tbird has raked up considerably more Pagefaults than a P3-733 box in roughly the same amount of time. My dual P3 at work currently has over 750 million page faults on each instance of the client!

I'm guessing this could be a good reason why the linux client is faster than the Windows one, this quantity of page faults must be cause a noticeable CPU overhead.

Scoofy12
04-16-2002, 10:18 AM
Originally posted by Dogbreath
I don't think so, a pipeline issue would be dealt with purely in hardware, it's not the kind of thing an OS would know about AFAIK.

You're right about that, pipeline misses should be a purely hardware thing. I'm guessing the foldtraj client generates quite a lot of those, id be willing to bet that i'ts very "branchy" and doesnt give a consistent data stream that is predictable by the pre-caching hardware. (after all, the pertinent data is generated pseudo-randomly). Perhaps this is why people are reporting better performance on Athlons.



I'm guessing this could be a good reason why the linux client is faster than the Windows one, this quantity of page faults must be cause a noticeable CPU overhead.

maybe... my linux clients show huge amounts of "system" cpu usage, about 20% (and 80% user, which is the foldtraj client itself) more than i've seen for any cpu-intensive process before. i think this reflects I/O and swap usage as well as memory management, scheduling and the like... maybe this is a reflection of the same phenomenon. I'm afraid my software knowledge is less than my hardware knowledge... does anyone know what an equivalent to page faults might be, or what this all might mean? i think it's been suggested that the client be less I/O intensive, but of course really only Howard can say what might or might not be beneficial. (of course, more user customization for these kinds of things is Always beneficial :D

iainmcgin
04-16-2002, 02:22 PM
Here's the definition of "Page Fault" as per the windows help files (an unlikely place to look for actual help! :p):


The interrupt that occurs when software attempts to read from or write to a virtual memory location that is marked not present.

In Task Manager, page fault is the number of times data has to be retrieved from disk for a process because it was not found in memory. The page fault value accumulates from the time the process started.

This perhaps makes sense if the foldtraj client is making quite a lot of use of disk caching as opposed to storing everything in RAM, however why the page faults? Surely such a standard procedure as reading and writing data to files doesn't constitute a page fault, per say?

I shall do some research :rolleyes:

Iain

bwkaz
04-18-2002, 12:58 PM
Nah, page faults only apply to RAM accesses, not file reads/writes.

For some reason, Windows is deciding to swap out a good chunk of the client's memory space, so then when the client accesses that memory again, the pages are marked not "present". So Windows has to swap them back in, and mark up a page fault.

But seriously, that many per minute? Wow, that's crazy...