SMP FAH on VMware

Maybe a bump wasnt needed but I thought I'd share a tidbit I came across with SMP VMware and 4 cores (be it dual dual or single quad) That I dont think is mentioned here, but could be "common sense" for most people.

Remember to set you affinity's in windows, dont let 2 VM's both share 4 cores, seperate them to each only use 2 cores. In testing my QX6700 overclock I had my machine freeze/reboot on me often, and a few times after a stable OC I fired up VMware and checked the frame times.... a few of those times I forgot to seperate the affinity's, here's the difference:

Protein 3025 (both vm's different run/clone/gen)
VM1(spread over 4 cores) 6:43
VM1(affinity set for 2 cores) 6:29
VM2(spread over 4 cores) 6:56
VM2(affinity set for 2 cores) 5:34

only 14s saved per frame on VM1, but VM2 saved 1:22 PER FRAME thats huge (near 25%)

Sorry if this is common knowledge and I seem stupid for stating the obvious but perhaps this will help a few !?!
 
Merom is complete plug and play for yonah if you don't mind swapping chips on a notebook. That'll get you 64 bit.

This may require a bios update and thats hard to get for most laptops.



 
So, after some grueling labor (as Oldbenwa can attest) I've finished updating this thing. It's currently moving from Ben's machine to mine via BitTorrent; feel free to join the torrent, it's designed for this kind of thing ;)

Improvements: Now uses a "getfold" script to get the latest version of FAH, so you can upgrade when a new version comes out. No guarantees about v6, though :p It's also ~10 megs smaller than the last version :D

So, to upgrade to the latest technical marvel, check out The Pirate Bay and join the torrent. HTTP mirror will be up as soon as I finish getting it here and putting it there.
 
This may require a bios update and thats hard to get for most laptops.

I wouldn't say most laptops, maybe some but certainly not most. Every dell lattitude and precision mobile workstation it is complete plug and play for and the same is true with all the lenovo thinkpads as they have all had bios's released to support it. Personally, those are the only two brands and series of laptops I buy or work with.
 
I checked my headless server the other day on which I am running the VMWare client and it had accumulated 221hours of idle CPU time while using 700hours for F@H. Would it make sense to add a regular F@H client to the mix, in order to suck up the last idle time? I had tried that before and it appeared to severely limit the VM's performance: 10 mins/ frame instead of 7. This is on a C2D, 2GiB with Win2k3 x64.
 
Adding another client causes a lot of context switches - the OS has to save the state of the VM, put the state of the single client in, and start the single client.
 
I checked my headless server the other day on which I am running the VMWare client and it had accumulated 221hours of idle CPU time while using 700hours for F@H. Would it make sense to add a regular F@H client to the mix, in order to suck up the last idle time? I had tried that before and it appeared to severely limit the VM's performance: 10 mins/ frame instead of 7. This is on a C2D, 2GiB with Win2k3 x64.

On my Dual-Dual Opty running Linux64 the SMP client only shows up to be useing 70% of the cores.
1 Core @100% & 3 cores at ~60-65%.
On my Dual Opty they show up as half the above.
I tried adding a second client to my dual-dual box but just slowed both clients down to half speed so no points gain.

Luck ........... :D

 
On my Dual-Dual Opty running Linux64 the SMP client only shows up to be useing 70% of the cores.
1 Core @100% & 3 cores at ~60-65%.
On my Dual Opty they show up as half the above.
I tried adding a second client to my dual-dual box but just slowed both clients down to half speed so no points gain.

Luck ........... :D

What about setting SMP to low priority (nice level higher than uni cpu) and the uni cpu client to idle
 
Got the SMP client up and running last night on a C2D 6400 @ stock 2.13..nothing else running that I can tell and I am pushing ~23 min per frame. I have 1GB of RAM dedicated to the VM (out of 2GB on the PC)..Do these #s look right? I keep seeing folks talking about 7 and 8 min per frame and I am just not seeing it. I have no idea what protein I am crunching (and if I could figure out how to do anything in VMware except let it fold that would be nice (like look at a log etc not actually use it) so I cant provide that info. I just hope I dont pass timeline on it. I will let it do its thing on this one to see if it finishes in time (it may be a big WU so I dont know) and maybe one more if this one fails as I was playing with VMWare for the 1st couple hours it was folding but if I keep getting out of deadline (and I am not saying even this one will it just seems like it is taking a while) then back to standard boxen.

BTW I find it funny that Tigerbiten outperforms me turning in his 20WU a day vs the 30WU a day I produce :)


 
Got the SMP client up and running last night on a C2D 6400 @ stock 2.13..nothing else running that I can tell and I am pushing ~23 min per frame. I have 1GB of RAM dedicated to the VM (out of 2GB on the PC)..Do these #s look right? I keep seeing folks talking about 7 and 8 min per frame and I am just not seeing it. I have no idea what protein I am crunching (and if I could figure out how to do anything in VMware except let it fold that would be nice (like look at a log etc not actually use it) so I cant provide that info.

You probably have a 2605 WU, 1760pts, 4 day deadline. Takes about 17mins/frame on my [email protected].
 
From the Folding-community support forum .............

CPU TYPE___|_Minimum Speed per core_|_Minimum VMware Speed per core
-----------+------------------------+------------------------------
A64 X2_____|_1.44GHz________________|_1.8GHz
Opteron____|_1.44GHz________________|_1.8GHz
Core 2 Duo_|_1.26GHz________________|_1.575GHz
Core Duo___|_1.75GHz________________|_N/A
Pentium D__|_2.555GHz_______________|_N/A
-----------+------------------------+------------------------------

So at 2.13 you should only take around 3/4 ofthe deadline times.

ps. I now get ~4k PpD from GPU folding, ~3k PpD from SMP folding, ~1k PpD from old fashioned CPU folding.

Luck ........... :D
 
Well if they would release a NVidia core I would be all over GPU folding. But as it sits the SLIed 8800s sit forlornly in the box along with the PS3 awaiting a client
 
From the Folding-community support forum .............

CPU TYPE___|_Minimum Speed per core_|_Minimum VMware Speed per core
-----------+------------------------+------------------------------
Core 2 Duo_|_1.26GHz________________|_1.575GHz
-----------+------------------------+------------------------------

So at 2.13 you should only take around 3/4 ofthe deadline times.
I must say that I am confused:
Got the SMP client up and running last night on a C2D 6400 @ stock 2.13..nothing else running that I can tell and I am pushing ~23 min per frame.
He should be able to meet every deadline.
 
Well if they would release a NVidia core I would be all over GPU folding. But as it sits the SLIed 8800s sit forlornly in the box along with the PS3 awaiting a client

Install Linux on your PS3 and put in the SMP Client there... maybe it would work or not. But put a linux client there... You know you can do this right? Who needs PS3 for gaming? :eek:
 
Install Linux on your PS3 and put in the SMP Client there... maybe it would work or not. But put a linux client there... You know you can do this right? Who needs PS3 for gaming? :eek:

Well since I literally have had it almost a month and haven't ever even turned it on I guess I dont need it for gaming :) And I am pretty sure that Zim01 tried the whole Install Linux and run folding on it and either it bombed or it ran like ass because he sold the PS3 about 2 days later.. I actually got the PS3 to watch movies on just been to busy working and watching HD-DVDs to give the BD movies a shake (and not much on the Blu side I care about)
 
Has anyone had any proper luck running this on Vista?

I'm running RC1 - 64bit, and I can get VMWare Server to install when I disable driver signature enforcement, and it will run until I reboot the machine. Upon reboot, I loose the "local server" option and I have to reinstall VMWare to get it running....

Anyone have any better luck than I?

I think the key issue is getting the driver to stick on the reboot for the virtual drive.

VMWare seems pretty adamant that their program won’t run in Vista and so does MS. So, I installed it anyway.

Can’t tell you a thing about 64bit because for the moment I have chosen to stay with 32bit ultimate, I might switch when more drivers become available.

That said I did everything from scratch, so maybe this will help.

During the whole VM install just keep ignoring all the windows incompatibility messages and continue on.

First I’m using version 1.01 of VM. Couple things I learned as I went along such as DO NOT allow Vista to disable auto start of your optical drive. In the event you forget that part you can delete if from your inventory later. The reason for this is VM won’t give rights to the optical drive back to windows, ever.

I used SUSE 10.1 64 bit and did a normal install. If you dl SUSE it will be a DVD download and contain 6 disks worth of material. During install you are given the option of adding other components or not, choose not and it will only use the first of the 6 disks.

You can also just use the ISO file in a virtual drive and allow it to install from there, VM can create the virtual disk for that. What I did was add the optical drive to my inventory for the install then deleted it.

When VM asks for a network connection when you start it you can use NAT or the shared / joined connections. Those would be the first two choices on the list.

Trying to install VM in compatibility mode simply seems to have no effect so I stopped trying to use it.

I tried this with both Unubtu and SUSE 10.1 both 64 and they seem to run just fine with the exception of, don’t even think about trying to make the computer clock and the Linux clock sync. It’s just not worth the pain. During config simply tell the client to ignore deadlines and that seems to work fine.

Between the two distros I very much preferred SUSE with the KDE interface.

Hope I made that clear, it’s late. Any questions feel free to PM me and I’ll do my best.

Luck, fold on.:)





 
Its weird, I downloaded VMWare, and the image. Running on a Dell WS that has 2x Dual core Xeons, that DO support virtualization and x64 (checked both in bios)

Got it up and running, configured it. Set affinity, and in Windows Task manager it is showing 2 of my 4 physical CPU's as being loaded at 100%. (Currently only running 1 VM)

Yet its been running almost 45 min now, without completing any steps. It is just chilling at 0 of 500000 steps.

???

 
Got it up and running, configured it. Set affinity, and in Windows Task manager it is showing 2 of my 4 physical CPU's as being loaded at 100%.

Yet its been running almost 45 min now, without completing any steps. It is just chilling at 0 of 500000 steps.

???



the free VMware server only supports 2 processors per virtual machine, wiht dual dual core you'll need to setup 2 virtual machines, doesnt sound like you've done that by what you said above. Then you set the affinity's so that each VM uses 2 cores and not windows default of spreading both Vm's over all 4 cores.
 
yes 2 virtual machines now each with affinity set to 2 DIFFERENT cpu cores. My origional post was made with only 1 VM set up.

However now that I have 2 VM's set up with manually assigned affinity, it is still taking about an hour to step.


1 VM is using core 0 & 1
the other VM is using core 2&3

It finally rolled over to 5000 of 500000 after 1 hour.
And these are 3ghz xeons on a machine with 2 gigs of physical ram.



 
It finally rolled over to 5000 of 500000 after 1 hour.
And these are 3ghz xeons on a machine with 2 gigs of physical ram.

Did the SSE boost start ok ?
If not they will run around 1/2 speed.
If its not on the stop the client & try restarting it with the "-forceasm" swith added.

Luck .......... :D
 
Yes SSE Boast is rolling. And both clients already have -forceasm.
I watch the little reminder about stability roll by when I start them.
 
anyone seeding the torrent?
I will be seeding this one until a new version is ready. I'm on a (usually) fairly fast connection, and I have seeded a little over 10 copies so far myself.
It finally rolled over to 5000 of 500000 after 1 hour.

Ouch! Is anyone else experiencing painfully bad frame times? Just as an experiment, can you run only one VM and put it on cores 0 and 2? I'm wondering if HT might be an issue on this machine - does it have (two sets of) two real cores or two cores with HT each?
 
I am getting about 23 min per frame on the 1760pt WUs (the only thing I have pulled so far in VM) Have seen no issues with the exception of that VM sucks to have running on non dedicated folding boxes (I am running this on my main PC) as it makes everything CREEP which I dont have the problem with the standard client. Yes, I understand it is because the VM wont release the cycles like the standard client does. I am just ready for the windows native SMP client to show up :)


 
Alright what happens when it pauses for 2min and can't detect the eth0 interface? Does that mean it can't load a driver/module for my NIC?

EDIT:nvm i fixed it with a intel pro 10/100 card
 
It has 4 physical cores. 2 sockets at 2 cpu's each. HT is turned off in the bios.

I tried various combinations of affinity etc, and finally deleted everything, uninstalled VMware and I started from scratch.

I am now getting about 15min/frame on 3ghz Xeon with 2 instances running. It still seems a bit slow compared to what people have been talking...but its a lot faster than 4x single clients.

 
hi guys,

I've started folding a few weeks back and i really love it.. its almost like an addiction..:D

anyway i like to try folding using SMP and Vmware but im having a hard time.. i already read this thread from the beginning and still cant make it work.. im a complete NOOB in using linux.. actually this is my first try..

ive followed this steps and still no go.. can anyone please teach me the step by step on how to configure this..

Code:
cd /home/fold
rm FAH_SMP_Linux.tgz
wget http://www.stanford.edu/~kasson/folding/FAH_SMP_Linux.tgz
gunzip FAH_SMP_Linux.tgx|tar -xvf -
./startfah

this is the error i got.

vmware.jpg


thanks!


BTW sorry if my english sucks.:p
 
this is the error i got.
Are you using the latest version? It'll be in a folder called smp-fah3... If so, run "getfold" from the command line and it'll download and unpack the latest version by itself. Running the latest version wouldn't hurt - it's only a hundred megs to download, and it does some convenient things like download FAH by itself.

If you want to use that set of steps to download FAH, you'll need to use the url http://folding.stanford.edu/release/FAH_SMP_Linux.tgz . Then make sure to use the same filename (ending in .tgz not .tgx) when you run the "tar" command. The reason it's changed is they put it on a different download server.
 
Thanks for the info and legwork on this mage, I just turned in a 1760pointer and I can do one of these a day. :) My [email protected] is doing about 14min/frame on the SMP client. It's a huge boost over just 2 windows clients.
 
I just wish they would release the Windows version already. (or the G80 version or the PS3 version as I have them all waiting). Need the Win version so I can use it on borgs since the VM doesn't release the cycles like the standard client does which makes it a no go for anything that I have to put into the field
 
Heh, now one of the 2 SMP instances, is stuck unable to upload, and of course it is at work, so very slow to work on. The other instance is 8 percent done, and has been running almost 48 hours.

My machine at home opty 170 @ 2.7 is getting about 18 min per frame. Does this sound about right?


 
My machine at home opty 170 @ 2.7 is getting about 18 min per frame. Does this sound about right?

That depends on the WU it is crunching. If it is a p2605, that's pretty quick. If it is a p3025 it is way slow.
 
Its a 2605, so 18 min is pretty good then.

Well thats good, now to get the work computer figured out!

Thanks for that info!

 
To start with, Thanks to all who put this together and thanks for the tech support in this thread, I needed it:D

Just finished a 2605 and cashed in big time,

The program worked like a charm on the rig listed below. My times looked like around 8mins a frame but this seemed to be a little off. I wonder if the clock is running a little slow.
 
Allright, sorry for being a linux noob here but here goes!



Still working on the aformentioned Quad Core Xeon.

Got 2 instances up and running. Only one is getting reasonable time frames, the other is taking almost 1.5 hours per frame.

Now, for some reason or another, the instance that takes forever to fold, uploaded grabbed a new unit (crunchy 1700ish) and is sitting at "Loaded Queue Successfully"
However it is throwing an error about unable to upload work unit.


The other instance is throwing some error about unable to upload work. It has 2 units in queue, and just downloaded a new work unit, and has started crunching.

I'm not sure what is different here, but I got SMP over VMware running on my opty at home just fine. But this damn xeon at work is driving me batty.

I have tried setting up a new directory for the stuff on the Xeons and whatnot, but they have just been weird and buggy. I am tempted to copy the
WORKING directory from my Opty, and just run it on the damn xeon.


I have no clue why 1 would work and the other wouldent. Any ideas?


 
hi unhappy_mage,

thanks for the reply.

i think i've found the problem.. i configured my network adapter to NAT instead of Bridge since i'm directly connected to my cable modem...

everythings working now..

thanks again.

regards
 
Back
Top