What affect do "page faults" have on SETI efficiency and what about future architectures and clients?

[JonB]

Woo. that's the longest title I've ever typed.

Anyway, looking at Task Manager, I noticed once again the huge number of Page Faults generated as the CLI client runs. I haven't checked 3.03 vs 3.08 vs GUI, but wonder if anyone else has.

I know page faults have to do with programs moving across memory boundaries, so a memory intensive program like SETI is prone to that, but it leads to some questions for the future.

1. If a program minimized page faults, would it run faster? There seems to be a lot of programming overhead coping with page faults.

2. With new CPUs and chipsets and larger memory addressing, will this reduce page faults? Do 64 bit busses help, how about 128, or 256?

3. How is BOINC about page faults?

I'm getting 21,400 page faults per minute of CPU time for SETI (3.03) for my AMD laptop. (WinXP Home)

JonB

 

[mechBgon]

I wonder if anyone could report what their page fault rate is with a large-cache CPU, such as a 1Mb Xeon? JonB, my Win2000 Task Manager doesn't seem to tell me the page-fault rate... what OS is that on, WinXP?

 

[JonB]

The Win2k or XP task manager will report Page Faults if you enable it.

Click on View, then Select Columns. Lots of choices.

 

[mechBgon]

Originally posted by: JonB
The Win2k or XP task manager will report Page Faults if you enable it.

Click on View, then Select Columns. Lots of choices.

Cool, /me learned something new! 😎

I'm getting 6,286,xxx page faults (per minute?) with the 3.08 client, on a system with 512Mb of PC2100 and an AthlonXP 1600+.

 

[JonB]

I'm getting 6,286,xxx page faults (per minute?) with the 3.08 client

At what % complete was your WU?

At the 48%, my WU had generated 3.4million page faults, giving me an estimated 7.13 million page faults when complete.

(divide page faults by percent complete, then multiply by 100)

I have an XP 1600+ also, but SDRAM laptop, 128 mb (96 RAM, 32 video)

 

[mechBgon]

Ahhh, ok... My present WU is 39.86% complete and has generated about 2,900,000 page faults. Scaling that up from 39.86 to 100, I'd be at around 7,300,000 when complete. For a 0.417WU I get times of around 3:48, so that's about 32000 page faults per minute.

 

[Assimilator1]

Bear in mind that v3.08 is about 12% slower/WU ,at least for average AR WUs.
For VLAR WUs 3.08 is faster,about the same speed for VHAR WUs

 

[JonB]

Ah, Assimilator-san, you see my question.
I want to know if 3.08 generates more or less page faults than version 3.03.

Mech is getting over 7million, and so am I.

I remember the glory days of assembly language programming when memory boundaries were respected and feared and page faults were avoided like the plague. Of course, a page of memory was pretty small back in the 8-bit days.

Now, is anyone running SETI on a 64-bit processor? and how does it do for page fault generation?

 

[Assimilator1]

Yeah I knew you wanted to know the page faults of 3.03 vs 3.08 but does it matter if 3.03 is faster?😉


Hey Mech!🙂
Long time no see in UT🙁
(mind you Cat doesn't often setup my server but likes to join the NOC ones)