Guide: Setting up notfred in VMWare Player

[nitrobass24]

Is anyone having problems getting Notfreds to reconfigure?
When I do it, it just sets the settings back to default.

WTF?


EDIT: FAIL
requires reboot

 

[baobab]

I think that the WinAFC portion should work fine if you have 2 VMs and 1 GPU.

I'm getting weird results running 2 VMs and 2 GPUs.

for me GPU0 has more shaders so it should outfold the GPU1.
after GPU0 finished a WU and loaded the next one, GPU1 would outfold GPU0. Until GPU1 finished or until I restarted GPU0 and GPU1.

right now I'm trying the following to attempt to overcome this.
GPU0 is in FAHGPU1
GOU1 is in FAHGPU2

*\vmware-vmx.exe := PAIR0+PAIR1 [assign=2,priority=Idle]
C:\Folding@home\FAHGPU1\FahCore_11.exe := PAIR0 [priority=High]
C:\Folding@home\FAHGPU1\FahCore_14.exe := PAIR0 [priority=High]
C:\Folding@home\FAHGPU2\FahCore_11.exe := PAIR1 [priority=High]
C:\Folding@home\FAHGPU2\FahCore_14.exe := PAIR1 [priority=High]

If someone has an input on how else to over come this, throw me a line.

I've manually split my affinities for the 2 VMs.

Your profiles look fine. In theory it should work as expected if you had your GPU clients' priorities set higher than the priorities of the VMs. But other people reported sort of similar issues with the GPU clients when ran as services.
http://foldingforum.org/viewtopic.php?f=52&t=7667
I do not know if you run them as a service or not. Anyway, those guys noticed that they had to use realtime priority for the GPU client to max their points.

To be honest, it does not make sense to me. I have a lot more experience with Unix than with Windows. If the kernel respected those priorities, than a process set at a higher priority level should always execute before any process with a lower priority. Maybe Windows implements this strict policy only for the realtime priority. I do not know. And it still does not explain why it would work as expected with only one client, or if the client was not a service (as per the linked thread).

But you can try the realtime priority for the GPU clients. You need WinAFC 0.9.2 for that and you must enable realtime priorities from Settings->General Settings.

I think I'm giving up on WinAFC.

It does nothing but slow my GPU clients down (at least as far as I can tell).
Without it, I have running 2xGPU and 2xVmWare clients on the rig in my sig, pumping just over 17KPPD.
With it, same config 14KPPD.

WinAFC just sets the process affinity mask for the processes you specify. The same as if you set the affinity in Task Manager. There should really be no difference between any affinity setting programs, except for what type of configurations you can set. But the affinity can be set in only one way using the Windows API. So you can check the relevant processes in Task Manager to see what affinity and priorities they have. You can always specify the same configuration in WinAFC and it should give exactly the same results. But of course, if the best you can do with custom affinities/priorities is the same as if you do not do anything, there is no point in using an affinity changing program.

I cannot get WinAFC to work for me.

I just copy and pasted from the guide. Maybe there is more to it.
It tells me that

I have an i7 920

The CPU detection heuristic is not reliable for the i7. I am surprised that it works 50% of the time. I just measure the communication bandwidth between all pairs of cores, and based on those results I determine if any set of cores have a higher affinity for each other (what I mean is if they communicate faster with each other than with other cores). The detection does not know anything about your hardware, except how many cores you have. It does not know if you have an Intel, an AMD or a VIA cpu.

Anyway, I wrote that benchmark with the Intel Core 2 Quads in mind, so I made some decisions at that point which proved to not be fully portable to the cache architecture of the i7. I am planning to rewrite the detection benchmark, but I will not have time to work on it until middle of May perhaps.

The post is getting fairly long, but I think for the i7 you may not need to run the detection heuristic based on the numbers that I've seen in the past, including yours.
I notice that the PAIRs of virtual cores that share a physical core are always (CPU0 + CPU1), (CPU2 + CPU3), (CPU4 + CPU5) and (CPU6 + CPU7).
Unless any of you that own an i7 got different results at some point, you can always use these physical identifiers in your affinity profiles: CPU0+CPU1 instead of PAIR0, CPU2+CPU3 instead of PAIR1 and so on. And you can disable the detection heuristic from the Settings menu.

If you do not understand what I mean, I can explain this part a bit more, just let me know.

EDIT: I restarted and it fixed itself. Not sure what that was all about.

Also, if the detection fails, or if you set it not to run at start, you can start it or re-run it explicitly at any time from the pop-up menu. It should be the 3rd option in there. The option is available only if the detection did not complete successfully yet. You do not need to restart the program.
Edit: Although you may need to restart WinAFC from the pop-up menu as well if it paused itself when not finding any valid profiles. So perhaps restarting the program is easier in the end

 

[capreppy]

Nitrobass, the error you received intially was because WinAFC had tried to determine your CPU and that check failed. Usually it fails because other processes were running that prevented it from doing so. This usually happen if you start WinAFC after your system has been up for a while. As you've seen a reboot normally corrects the problem.

 

[Wesleynator]

Capreppy - things are working better. I have it running seeminly stable on 2 out of 3 boxen now. On the 3rd machine notfred's dectects my cpu as i686 not x86_64 so it loads up a classic client instead of smp. This is on a Q6600 like all my other machines. I have tried uninstalling and reinstalling 3 times and it turns out the same. Anyone have any idea why? Many thanks for fielding all these support questions. Write a guide to help a few folks and you turn out to be support for the world...ha..

 

[capreppy]

Only reason it would detect i686 and not smp is the "folding.vmx" is showing 1 cpu instead of 2. Check the file to make sure it is set to 2 cpus.

I actually don't have an issue with supporting everyone. We're all in this together and when it comes to CC '09, I'd like us to have a fighting chance

 

[Wesleynator]

One more question...how do you use fahmon to monitor these vm's on other machines on the local network? I tried using \\machinename\fold-xxxx\C\etc\folding\1\ also \\fold-xxx-C\etc\folding\1\ but no luck. Thanks, guys.

 

[capreppy]

One more question...how do you use fahmon to monitor these vm's on other machines on the local network? I tried using \\machinename\fold-xxxx\C\etc\folding\1\ also \\fold-xxx-C\etc\folding\1\ but no luck. Thanks, guys.

I use (per the guide) \\fold-xxxx\C\etc\folding\1\ I use this for all of mine using either FahMon or FahSpy. This is also the method recommended by notfred's on his website. Check to make sure that it is as I have it above. Let me know if this does not correct your issue.

 

[Wesleynator]

Yeah, mine looks just like yours. I am able to monitor a vm on the same machine using that syntax but I want to monitor all my boxen from one machine and this is where I'm having trouble. I can put that syntax into fahmon on my main box for a vm on another machine and it just says something is wrong with the client. I can monitor the gpu and winSMP instances on the remote box no problem. I must be leaving something out.

 

[capreppy]

You may want to change the workgroup on the VM instances to be the same as your monitoring machine. See if that corrects it. All of my VMs are configured to be in the same workgroup as my main box. I've not actually done an instance using the default 'DISKLESS".

 

[Wesleynator]

That's a good idea but I already made sure that matched. If you click on Network do your vm's show up along with the other pc's on your network?

 

[Vaulter98c]

im mobile now, so I cant check my box, and I had to shut down on 96 do to PSU power issues ( RMA should be back tuesday tho ) but my VM's have been running stable with that one exception. I'm still trying to figure out the overall loss of PPD on my cards (all cards are down 1k on avg, 1.5 on the 88GT, but the 88GTX is 1k over previous PPD) but the VM's are doing fine, running on 2 mobo's, both using Win7

 

[nitrobass24]

Anyway to run VMwarePlayer in the background like as a service so I dont have to manually fire up 4 of these everytime, and have them taking up room on my taskbar?

 

[capreppy]

That's a good idea but I already made sure that matched. If you click on Network do your vm's show up along with the other pc's on your network?

Yes, I can see all of my VMs when I look in my network.

Anyway to run VMwarePlayer in the background like as a service so I dont have to manually fire up 4 of these everytime, and have them taking up room on my taskbar?

You can create a "startup" task in Task Scheduler to start them up at startup of the box and they'll be hidden. I did try this, but didn't like it. The issue is if you need to kill it (to play games or whatever), you'll not have visibility to it. You could kill it through Task Manager by killng the vmware-vmx.exe file. There may be a way to set it as a service, but I'm not that familiar with it to make suggestions.

 

[osrk]

Is a linux vm install more stable then a notfred install? ie which would provide more points and more stability?

 

[Zero82z]

Is a linux vm install more stable then a notfred install? ie which would provide more points and more stability?

Doesn't really make a difference. I did get some out of memory errors while running notfred's though, so I went back to my previous standard Linux VM. I could probably get notfred's up and running again if I bothered to tweak it a little more though.

 

[nitrobass24]

You can create a "startup" task in Task Scheduler to start them up at startup of the box and they'll be hidden. I did try this, but didn't like it. The issue is if you need to kill it (to play games or whatever), you'll not have visibility to it. You could kill it through Task Manager by killng the vmware-vmx.exe file. There may be a way to set it as a service, but I'm not that familiar with it to make suggestions.

Yea but when you boot notfreds you have to tell it what to do,
eg. Fold, fold64, bench etc.

 

[Zero82z]

Yea but when you boot notfreds you have to tell it what to do,
eg. Fold, fold64, bench etc.

It will automatically pick one after a certain period of time though. If you have it set up properly, it should default to fold64.

 

[Tigerbiten]

Anyway to run VMwarePlayer in the background like as a service so I dont have to manually fire up 4 of these everytime, and have them taking up room on my taskbar?

I'm useing VMWare Server v1.??
I've just set it up to start the VM's at host power on.
This starts the VM in the background, then the SMP client starts inside it as a service.

Never played around with VMWare Player but there should be a setting to start it at host startup as well.

Luck ............

 

[capreppy]

Unfortuntaely there aren't many options in VMWare Player. All of the startup functionality is in VMWare Server. I haven't noticed any stability difference between VMWare Player & VMWare Server. VMWare Server seems to be a bit more finicky though

notfred's will choose an option 15 seconds after startup. It will choose either fold64 or fold. If you've been able to run fold64, it will choose fold64

 

[Kendrak]

Just a quick question.....

Do the VMs give up cpu cycles if they are needed? (On my main rig I'm debating running 1 or 2 of these. Q6600 with a good OC)

Also, does the VM save the progress?

Sorry for all I've folded I've never played with VMs before.

 

[Vaulter98c]

I run 2 on my Q at stock and it's good enough for web browsing, streaming music, running torrents, burning cd's, dl'ing shit in firefox and monitoring my foldi g and all temps inrealtime, only had a fe issues with stuttering the Internet radio, everythng else wasfine, was a real busy day today

And it's running 2 gpu clients too, all the same rig

 

[Vaulter98c]

And yes, it saves progress, you can set the interval, I put mine at 5min, I think that's the best

Sorry for the double, i'm on my phone

 

[capreppy]

Vaulter98c is my case study

 

[Vaulter98c]

I try

 

[Wesleynator]

I waved the white flag last night and went back to the Windows SMP client. I really wanted this to work out but I could not complete 2 units in a row without an error, typically the one about attempt to access beyond end of device. The GPU client on one of my boxen also took a ppd hit, sometimes enough to negate the extra points from the vm smp. I know a lot of you guys have made this work which is why you are the elite folders for the Horde. If there are any new releases or improvements, shout it out and I'm always game to try again.

 

[capreppy]

I am very sorry it did not work out for you. I will definitely keep you and others appraised if notfred has an update.

 

[EvilAlchemist]

Itypically the one about attempt to access beyond end of device.

I get this error if I use the VM files provided on the site.

If you use the ISO from the site this error will go away.
You will loose your stop/ start functions (reboot of VM --> new WU) but that is not an issue in my setups.

Also, I would SUGGEST increasing the ram per VM.640 MB is not really enough IMO.

Since All my system has 4GB+ of ram, I run each VM instance with a minimum of 1.5GB RAM per VM and mine have been running stable for 82 days now.
Some of the Large SMP units have been known to use a ton of memory.

 

[capreppy]

EA, I'll update the OP to make that suggestion

 

[nitrobass24]

I get this error if I use the VM files provided on the site.

If you use the ISO from the site this error will go away.
You will loose your stop/ start functions (reboot of VM --> new WU) but that is not an issue in my setups.

Also, I would SUGGEST increasing the ram per VM.640 MB is not really enough IMO.

Since All my system has 4GB+ of ram, I run each VM instance with a minimum of 1.5GB RAM per VM and mine have been running stable for 82 days now.
Some of the Large SMP units have been known to use a ton of memory.

Yea I have been seeing some out of memory errors on my vm's over the past few days.
I have since increased the ram to 1gb, and its been smooth sailing.

 

[Vaulter98c]

does that up PPD? or just stability?

 

[capreppy]

It only affects stability.

 

[nitrobass24]

It only affects stability.

QFT.
The only thing that can really up PPD is OC'in

 

[styx0r]

Just wanted to chime in with my own experience - if you reboot a client with VMplayer's Troubleshooting->Reset function, it will reset the affinity and priority of the client's vmware-vxm, and WinAFC will not catch this. I realized this after my SMP clients would report uneven PPD on (basically) identical work units.

Also, increasing the VM memory allocation, as recently suggested, has a very positive effect on stablity. Take your time, be patient with this method, and you'll be happy you did. It's powerful and easy to turn on and off.

Thanks again capreppy!

 

[Vaulter98c]

what would you suggest, 1 gig each? Ive only got 4 total to play with and the system needs 1.5 atleast

 

[capreppy]

Just wanted to chime in with my own experience - if you reboot a client with VMplayer's Troubleshooting->Reset function, it will reset the affinity and priority of the client's vmware-vxm, and WinAFC will not catch this. I realized this after my SMP clients would report uneven PPD on (basically) identical work units.

Also, increasing the VM memory allocation, as recently suggested, has a very positive effect on stablity. Take your time, be patient with this method, and you'll be happy you did. It's powerful and easy to turn on and off.

Thanks again capreppy!

Styx0r, thank you for the feedback. I've never played with the Reset function as I always restart th VM via the Web UI for notfred's. This is definitely good information.

what would you suggest, 1 gig each? Ive only got 4 total to play with and the system needs 1.5 atleast

1GB each should be sufficient. On my i7, I'm running 12GB so I give it a bit more.

 

[flogge]

Got a VM running on my sig machine along-side the 4890. May I say that this GPU at 945 Mhz is one thirsty critter for CPU cycles. I still have some OCing to do. The 720 is only at 3.2 so far and the GPU at 945. I need to get a few more system monitoring apps installed on my "HTPC" before I reach for the anti-gravity button.

 

[Nessus85]

Woot! Im at 12,521 ppd thanx to your vmware guide + two 8800gs's.

 

[Pocatello]

This should be linked in the Horde guides page.

 

[capreppy]

This should be linked in the Horde guides page.

Xilikon put it up there a week back or so.

 

[Pocatello]

Xilikon put it up there a week back or so.

Really?

I could not find it. I just went looking for it in all the easy places.