affinity changer

[Assimilator1]

From above linked thread:-

How does it work? <- new info

After some time interval (10 mins) utility checks, if there are 4 processes of FahCore_a1 (one FAH SMP client). If so, and system has 2 CPUs or 2 cores, then utility assigns two processes for each core in some magic way: first core gets process with maximal memory usage, and process with minimal memory usage. Other two processes are assigned to second core.

For quad core processors, first SMP client is assigned to 1st and 3rd core, 2nd SMP client assigned to 2nd and 4th core. Assigning affinities on QUAD if you run one SMP client does not make sense.

 

[GLeeM]

From my understanding:

It gives a nice increase in ppd by assigning affinity to F@H SMP cores instead of letting Windows do it.

Windows is designed to handle any number of processes and so affinity needs to be able to change on the fly.

F@H SMP WUs use lots of L2 cache well. The way the Intel Core 2 shares L2 cache per core and the way Windows assigns each of four processes per WU is not always the best. This program is trying to fix that.

Warning:
By using the program, each of the processes are fixed to a core.
If the computer needs to do lots of other work, like a very busy server or even has just one other process that uses one core all the time by more than say 20% CPU usage, then this program might not help get better ppd because the other three F@H processes would have to wait for the fourth to catch up! (20% is a wild guess on my part)

 

[GLeeM]

Originally posted by: GLeeM
As for his zealous friend, he might not understand about these things and only sees how much good nick4evas program does for F@H science!
It's hard to fault a friend for doing what he thinks is right.

<The thing that bothered me was that he thought asking about the program implied "security noob".
Which implied that the noob didn't know how to crack or reverse engineer a program to check for malware and that this practice is easy for the non-noob to do. A lot of people might think that behavior unsavory and thus make them wonder about security issues!

Just caught up on the thread over at F@H forums.

Boy am I embarrassed for the quoted first paragraph, but the second is proven true.

Nicholas needs a new PR (public relations) guy
Imagine his embarrassment!

 

[Insidious]

Thanks GLeeM!

I think you're right on the money with your last post!

What kinds of things can be done for ourselves help everyone be comfortable enough with this utility for GLeeM to add a link to it in the stats post he gives us daily?

For myself, I am fine using it, but you have to put that in the perspective that I am constantly backed up and don't really have information that would embarrass me or hurt me if it was taken or damaged.

I remember that Jim (caferace) and ScottV67 mentioned some process they could go through to make sure it was not 'phoning home or something' either of you guys got the time to mess with it?

-Sid

 

[biodoc]

At this point, IMHO it's safe to use. The developer is visible now and has provided the source code to the community. I'm not a security expert by any means and so far nod32, vista firewall nor my wireless router has shown anything in the logs.

I'm going to continue to use it because it works as advertised

More PPD!!!

:beer::beer:

 

[uncle fungus]

The source of the decompiled binary is essentially the same as that posted in google-code. I say essentially because decompilation never gives the same code back.

The code is safe though and doesn't do anything other than what it purports to do.

 

[Insidious]

Originally posted by: uncle fungus
The source of the decompiled binary is essentially the same as that posted in google-code. I say essentially because decompilation never gives the same code back.

The code is safe though and doesn't do anything other than what it purports to do.

I recognize this name!

WELCOME to the AnandTech forums Uncle Fungus!

I hope you will visit us often.

:beer:

 

[GLeeM]

Originally posted by: Insidious

Originally posted by: uncle fungus
The source of the decompiled binary is essentially the same as that posted in google-code. I say essentially because decompilation never gives the same code back.

The code is safe though and doesn't do anything other than what it purports to do.

I recognize this name!

WELCOME to the AnandTech forums Uncle Fungus!

I hope you will visit us often.

:beer:

Yes! I recognize it too! He does the fahinfo.org site

He is the best mod/admin over at the F@H forums

Thanks for the info :thumbsup:

 

[biodoc]

Originally posted by: GLeeM

Originally posted by: Insidious

Originally posted by: uncle fungus
The source of the decompiled binary is essentially the same as that posted in google-code. I say essentially because decompilation never gives the same code back.

The code is safe though and doesn't do anything other than what it purports to do.

I recognize this name!

WELCOME to the AnandTech forums Uncle Fungus!

I hope you will visit us often.

:beer:

Yes! I recognize it too! He does the fahinfo.org site

He is the best mod/admin over at the F@H forums

Thanks for the info :thumbsup:

Ditto!!:thumbsup:

Thanks for the visit & the info uncle fungus!!

 

[uncle fungus]

Thanks for the welcome guys. I'm very busy elsewhere and in real life at the moment, but I'll try to drop in every so often

 

[petrusbroder]

Thanks for coming by and welcome!

 

[Slider76]

Welcome Uncle fungus, its always good to see a dedicated person such as yourself join the forum.

 

[caferace]

Originally posted by: Slider76
Welcome Uncle fungus, its always good to see a dedicated person such as yourself join the forum.

Ditto. Though I keep trying to type Uncle Fester.

UF, thanks for your assistance on the FForum.org board.

-jim

 

[biodoc]

I really like that coiled snake icon

Maybe I'll switch mine!

:beer:

 

[Assimilator1]

Originally posted by: GLeeM
From my understanding:

It gives a nice increase in ppd by assigning affinity to F@H SMP cores instead of letting Windows do it.

Windows is designed to handle any number of processes and so affinity needs to be able to change on the fly.

F@H SMP WUs use lots of L2 cache well. The way the Intel Core 2 shares L2 cache per core and the way Windows assigns each of four processes per WU is not always the best. This program is trying to fix that.

Warning:
By using the program, each of the processes are fixed to a core.
If the computer needs to do lots of other work, like a very busy server or even has just one other process that uses one core all the time by more than say 20% CPU usage, then this program might not help get better ppd because the other three F@H processes would have to wait for the fourth to catch up! (20% is a wild guess on my part)

Interesting ,so if I were to game for 2-3hrs would this then negate the benefits gained from A Ch?

uncle fungus welcome to AT

Biodoc
You picked the icon from the old Gamma Flux DC project (RIP GF)

 

[GLeeM]

Originally posted by: Assimilator1
Interesting ,so if I were to game for 2-3hrs would this then negate the benefits gained from A Ch?

You would have to check for yourself. I made that statement from logic.

If I continue by logic: the game does not use all of a CPU during that 2-3 hours. Maybe one hours worth? I think overall A Ch helps more than that?

I have noticed that one of four processes has lots more I/O Reads and is probably the main process?

Another one of four processes has less CPU time used, and if I run another full time process, this one is the process that runs least, (I don't use A Ch). Even though Task Manager shows full use it seems otherwise. I wonder if this is the reason that running two SMP on a quad is advantageous?

So of the four - one seems to use more and one seems to use less, the other two seem pretty even. (Did I read somewhere that A Ch matches the most and least used processes to one core? This would make sense!)

All of above is mostly from my observations.

 

[Insidious]

I was really curious about this too.

My son (God love him!) bought me Orange Box for Christmas. He remembered how much fun I used to have with him on the original Team Fortress Classic when he was a kid.... so now I get to re-addict to Team Fortress Classic 2,

Anyway, the longest I usually play at a time is an hour or so and I went back to look at FahMon afterwards. Yep, one of the clients (I run 2 @ SMP) took a huge hit while I played.... but the other stayed pretty much normal. The slow one did bounce back to normal times when I was done with the game.

So I think it is more of an acedemic exercise they are pointing out that affinity changer does not even out the load of the game with the loads of the folding.

-Sid

edit: @ GLeeM: I think A Ch looks at the memory usage of each thread. Max and Min to one core, the two 'mediums' to the other. (also, for a quad... one SMP client goes to core 0 and 2, the other SMP client goes to core1 and 3)

 

[LOUISSSSS]

where can we download this affinity program? i can only get F@H to run on 1 of my 4 cores on my Q6600...

 

[Alyx]

Originally posted by: scottv67
caferace wrote:

And I'm guessing that person is a 17 year old that lives in his Aunts basement.

Jim, you know that's not true: I turned 18 a few weeks ago (and my aunt *does* charge me rent.)

-scott :beer: :beer: :beer:

does " *does* " make it safe to assume that her choice to charge rent does not necessitate getting payed rent?

Because thats much the same as not charging all together

Edit: I didn't realize that I just quoted a post that is a month old. Ohh well.

 

[caferace]

Originally posted by: Alyx
Edit: I didn't realize that I just quoted a post that is a month old. Ohh well.

:laugh:

LOUISSSSS, you can get it here: http://code.google.com/p/fah-smp-affinity-changer/

-jim

 

[Insidious]

Originally posted by: LOUISSSSS
where can we download this affinity program? i can only get F@H to run on 1 of my 4 cores on my Q6600...

The affinity changer is for the SMP client. when two instances are running at once.

It sounds like maybe you are running the no nonsense text only client (CLI)

It won't run on more than one core no matter what. You could install 4 instances of it if you want to use the whole Quad, or install the SMP client.

If you choose to use the SMP client, a single instance will use all 4 cores, but the affinity changer will not affect it.

The affinity changer is for when you run two instances of SMP on a quad.

-Sid

 

[mooseracing]

do you think it would be better running affinity changer on smp on my dual dual cores?

Or should I be running 2 smp clients and affinity changer?

 

[caferace]

Note that affinity changer is for Intel only. Doesn't do anything for AMD except slow things down (slightly).

-jim

 

[Insidious]

Originally posted by: mooseracing
do you think it would be better running affinity changer on smp on my dual dual cores?

Or should I be running 2 smp clients and affinity changer?

for Dual core machines, only run ONE instance of the SMP client (you need a minimum of 2 cores for each instance).

It's not really that the affinity changer is not for AMD... it is that the affinity changer is for quad core PCs. At the moment, that means Intel, but not forever (hopefully).

I run a single SMP on my X2 machines and the affinity changer didn't make any change that I could see at a casual glance.

-Sid

 

[caferace]

Originally posted by: Insidious

It's not really that the affinity changer is not for AMD... it is that the affinity changer is for quad core PCs. At the moment, that means Intel, but not forever (hopefully).

I run a single SMP on my X2 machines and the affinity changer didn't make any change that I could see at a casual glance.

-Sid

Over at Folding Forum.org, there was a bit of a revelation from Rilian re: AMD:

http://foldingforum.org/viewto...4&t=136&start=45#p4852

-jim