SMP folding guide with VMware player

[leexgx]

Hoping that this post will be helpful for windows users who want to fold using SMP but want to get full use out of their hardware when using CPU folding

Due to the way MPICH or Deino works it is very un-optimised for windows OS (I am thinking of Work done here) was going to stop them as my GPU2 clients did Far better but the link below at custom pc/bittech made me have an go at Linux in an Virtual Machine, I was very surprised once it started doing work using NotFred linux SMP in Virtual Machine (VM for short and will only type VM from Now on) is like running the 1 GPU client for some or some times like 2 for lucky i5 or i7 users at high clock speeds, you get far more work done

As most do not want to use Linux as there main OS you can use an OS called NotFred (there are others but this one is made Just for SMP only and meant to get up and running in Very short time) it made to be disk less so you can run it off an CD or an USB Pen drive or in an VM for the Task that i made this Guide for Running it in an VMware player program (supports 4 cores each VM) or VMware workstation 6.5.3 or newer(supports 8 cores each VM useful if you have an i7 with HT) so this means you do not have to mess with partitioning your hard disk or even need to learn how to use Linux

The new VMware player now that supports 4 cores (its free as well) so you no longer need to run 2 copy’s of VMware player to use 4 cores on an Quad CPU so less ram is now needed to run VM NotFred or your VM linux box, it is also far simpler to run it as well, with what I did below you can be up an running in 1 reboot and 5-10 mins and it serves the Folding@home project far better due to work units can be turned around far faster (up to 2x faster then the WinSMP until they bring an windows optimized SMP client but the Linux SMP may always be better)

http://www.overclock.net/overclock-net- ... layer.html

Make sure you have at least 2-4gb of ram and an 64bit OS is recommended, as the VM can use up to 1.2gb of ram (if your doing GPU folding as well you may need 4gb of ram most likely)

1.
Download the below
http://www.vmware.com/products/player/ or get VMware workstation 6.5.3 {tested} or higher (mite be needed for 8 core systems or i7 systems with 8 threads)
http://reilly.homeip.net/folding/vm.html (this is Notfred it supports 2-4-8 cores)
2.
And edit the folding.vmx file (3KB not the 20mb one) in windows 7 and 2003 server i had to use wordpad to edit my folding.vmx (due to notepad merging all the lines)
3.
change the below
numvcpus to 4 (allows 4 core to be addressed only on VMware player, if your useing i7 or 8 core system you can test 8 if VMware work station is been used)
memsize to 2000 (allows 2gb of ram to be used, the Default is to low and may crash)
add line at bottem (make an new line)
priority.ungrabbed = "idle"
Prevent slow downs more important if your doing GPU2 folding as well, as Norm priority {default} impacts on GPU2 progress speeds, you should not need to mess with GPU2 as i have not on 2 systems and they get the same PPD with or with out the VM running, when set to idle
4.
Start the Notfred VM (open the folding.vmx with Vmware player or workstation) if it works you see an ip address like 192 if that's your routers DHCP server IP (not 169. as that be failed to get IP address) it takes about 30 secs to norm boot to the login part (do not login not needed)
5.
To access notfred use Internet Explorer or Firefox web browser and type fold-xxxx into the address bar (not the search box) press enter or go (look at your VM notfred for the host name) if that fails login via the IP address (look at VM for IP Address) you must do this on your Windows PC (the VMware Notfred is ready when it gets to login part in the VM box, do not try and login its not needed, its done via web browser settings)
6.
Press Reconfigure this host and any USB drives and put your name that you want and your team name, put your pass key in if you want to, make sure SMP instance per CPU its set to 4 (i7 or 8 core systems try 8)Go to the bottom and press Reconfigure and then press {Remote reboot: enabled} button and when it starts back up, it start with your settings,
7.
If you need to power it down it just stop the VM make sure the disk light is not flashing on the VM or the backup it makes be broken, notfred makes an backups every 15 mins when the VM starts back up it restore that data so you lose at most 15 mins of work (norm 1-3% for most)
---------------------------------------------------------------------------
Fahmon setup
To monitor the program using program like Fahmon \\fold-xxxx\C\etc\folding\1\ , xxxx being the host name in the VM notfred (Look at the VM Notfred box)if that fails you need to use IP instead of fold-xxxx, unlike windows linux on restart will norm not always get the same ip address so the host name is the better option, you need to do the below if you want it static IP (you can just try setting the workgroup to match your systems workgroup on Vista and win7 its norm workgroup then you may be able to access the web page then)
(do below if your ok with setting IP address)
you understand how to use the reconfigure part of notfred, set the IP address manually (norm 192.168.x.201-250, between 201-250 must not match any other IP on your network) gateway point that at your router (norm 192.168.x.1 BT routers some times use 254) and DNS (192.168.x.1 norm points to the router, some point to the ISP DNS) the x is router Dependant most routers go between 0 and 1 belkin norm 2, you can check this by checking your ip address on your windows system (fast way start > run > CMD > type ipconfig /all press enter and look for the one with 192 in you see gateway and DNS there), once you done that you should be able to use the IP address all the time (fahmon add it as \\<IP-ADDRESS>\C\etc\folding\1\ )

---------------------------------------------------------------------------
As Stanford prefer to get work back as fast as it can it is better to use 1 VM Notfred at an time (as i found 2 VM are likely going take 2.5x longer to send the work back)

Code: Select all

i have some issues on my QX9850 the time clock in the VM looks like is going 1.5x slower when 2 VMs are running (clocks counting to fast), in FAHmon as it thinks the % times are faster then what it is i am guessing it may be slower to use 2 VMs then one only (i watch it in an bit to log the real time when its working), i recommend to only run 1 VM notfred

Ok do not run 2 notfred VMs as it takes a lot longer to complete each work unit 11hrs on one VM (thats 2 per day) on 2 VMs it takes 33hrs (thats about 2 per day and an half, so 20 hrs is wasted)

do not be fooled into that your getting high PPD when running 2 VMs, its FAHmon is been miss lead, in this case due to the fact that each VM is running at 2x slower Time speed so it says its 10 mins per frame but its really 20 mins real time but FAHmon thinks its 10 mins as its only looking whats in the folding logs
---------------------------- FOR I7 or 8 core users --------------------
if you have an I7 system or 8 core system you can get Vmware workstation 6.5.3 or newer to use all 8 thread, set it to use 8 threads via numvcpus to 8 in the fold.VMX (this is untested need feedback), , {need info on how to edit the init file so to make -bigadv work as i7/i5 with high clocks may want this}
----------------------------------------------------------------------------

[leexgx]

Reserved (will add pictures in the next week or so before dec 25th)

just putting this up for users who use GPU Folding as well as VMware-notfred (add more info in the short time)

VMware folding affects the NV GPU folding cores (FahCore_14.exe) , you need to make sure FahCore_14.exe is set to realtime using the program below (make sure the GPU cores are spread over muti cores like max 2 per CPU any more then 3 GPU cores per CPU can suffer PPD loss) FahCore_11.exe can be set to High or real time (does not affect the points, but the FahCore_14.exe needs to be set to real time using prifinitty or you get half ppd on the NV 14 core projects)

http://edgemeal.110mb.com/PriFinitty/index.htm

[toTOW]

I'm sorry you had to invent the wheel again on VMWare optimizations : http://en.fah-addict.net/articles/artic ... ations.php

[leexgx]

but its not on here on how to do it most would not look beyond folding@home fourms (i not checked that link as i am fixing computers at this time)

leagily you have to pay for VMware workstation so you would have to buy it , where as VMware player is free thats what is new as it now supports 4 cores so you can download it for free from VMware web site in the provided link and use the Notfred VM that is precofig for SMP 4 cores just need an little edit in the VM config to use 4 cores as it default is to use 2 when you open the VM

[toTOW]

Yes, I'll update my article to : "this tips works on Workstation 6.5x and 7, and on Player 3.0"

[geokilla]

Fail....did you really have to make a guide when there's tons out there?

[Mactin]

leexgx,
Thanks a million !
I was wondering on how to do this and had no time for research, your guide comes just in time for me.

[leexgx]

Fail....did you really have to make a guide when there's tons out there?

but its not posted on here thought (is now), if you can imporve it be help, going to edit it an little soon

most stuff i have seen guide wise is using ubuntu and it needs some sort of basic linux know how to set it up, what i have posted can be done in 10 mins, and its only basic linux and console and FAH SMP, the notfred is run and forget once you put the correct team and name in that you want to use and edited 3 lines in the VM settings file that takes like 20 secs to do (10 mins is more download time and Reconfiguring the notfred)

still why so negative posts what i posted here is very usefull for new users who are not l33t with linux and it gets the best out of there system in an short time with out haveing to move to linux completely or mess with commands that can be longer then my hand

@Mactin
how long did it take you to do it and did i miss any thing out

[Grandpa_01]

You need to change your VM player link to this one VM player 3 is available now.http://www.vmware.com/products/player/

[#64-Lux]

Many thanks for the guide! Just in time for me!

Small addition for total newbies like myself. I have an IP assigned by ISP and VMPlayer couldn't connect to Stanford by default.
I had to change VMPlayer -> VM -> Settings -> Network Adapter -> Network Connection -> NAT. (default was Bridged).

Thanks again, got a nice ~2000-PPD boost.

[weedacres]

Many thanks for the guide! Just in time for me!

Small addition for total newbies like myself. I have an IP assigned by ISP and VMPlayer couldn't connect to Stanford by default.
I had to change VMPlayer -> VM -> Settings -> Network Adapter -> Network Connection -> NAT. (default was Bridged).

Thanks again, got a nice ~2000-PPD boost.

If you have a router that's assigning LAN address's (DHCP) bridged should work just fine. Your VM should end up with an address on your LAN's subnet.
If you run NAT you'll be unable to use FAHMon or HFM.NET without a lot of jury rigging.

[uncle fuzzy]

Many thanks for the guide! Just in time for me!

Small addition for total newbies like myself. I have an IP assigned by ISP and VMPlayer couldn't connect to Stanford by default.
I had to change VMPlayer -> VM -> Settings -> Network Adapter -> Network Connection -> NAT. (default was Bridged).

Thanks again, got a nice ~2000-PPD boost.

If you have a router that's assigning LAN address's (DHCP) bridged should work just fine. Your VM should end up with an address on your LAN's subnet.
If you run NAT you'll be unable to use FAHMon or HFM.NET without a lot of jury rigging.

Mine are all NAT, wouldn't do a thing Bridged (Player and notfred's). All are monitored by FahMon, either using \\fold-xxxx\c\etc\folding\1 or the IP for the VM. It was only my one shot using Ubuntu instead of notfred that I couldn't monitor.

Mine are only monitored locally. If you're talking monitoring multiple machines over a network, that might be another story.

Contrary to the OP, before getting Player 3, I ran 2 VMs on my quads and got double the output. All I did was use Task Manager to set low priority and set each VM's affinity to a different pair of cores. This is using tear's MPICH fix. Without the fix, each VM was still faster than a single Win SMP.

1 Win SMP- ~24-30 hours
1 VM- ~ 18-20 hours
2 VM- ~ 18-20 hours

[weedacres]

Mine are only monitored locally. If you're talking monitoring multiple machines over a network, that might be another story.

Yes, that is what I'm talking about.
My guess is that if you can't make bridged work then you might have a firewall problem on the host. I have 5 VM's running bridged on XP and W7 machines with no problem.

[Grandpa_01]

[
Contrary to the OP, before getting Player 3, I ran 2 VMs on my quads and got double the output. All I did was use Task Manager to set low priority and set each VM's affinity to a different pair of cores. This is using tear's MPICH fix. Without the fix, each VM was still faster than a single Win SMP.

1 Win SMP- ~24-30 hours
1 VM- ~ 18-20 hours
2 VM- ~ 18-20 hours

Uncle fuzzy are you saying your WU's run slower with player 3 than they did with player 2 on your quads. I am running 4 quads 2 smp notfreds clients on each what I have seen so far is that my 2 Intel Q9650 are faster with player 3 and my AMD 9950 also gained a little, but my AMD 940 seems to be slower. They are all using tear's fix. I do set the priority but I do not set the affinity, I could not see where setting the affinity did me any good before.

[uncle fuzzy]

Contrary to the OP, before getting Player 3, I ran 2 VMs on my quads and got double the output. All I did was use Task Manager to set low priority and set each VM's affinity to a different pair of cores. This is using tear's MPICH fix. Without the fix, each VM was still faster than a single Win SMP.

1 Win SMP- ~24-30 hours
1 VM- ~ 18-20 hours
2 VM- ~ 18-20 hours

Uncle fuzzy are you saying your WU's run slower with player 3 than they did with player 2 on your quads. I am running 4 quads 2 smp notfreds clients on each what I have seen so far is that my 2 Intel Q9650 are faster with player 3 and my AMD 9950 also gained a little, but my AMD 940 seems to be slower. They are all using tear's fix. I do set the priority but I do not set the affinity, I could not see where setting the affinity did me any good before.

No, once I got Player 3 I could run 4-core SMPs. I've changed 2 of them (Q9550 and OC'd Q6600) to Player 3, and 4-core is the same with, or without, the fix. The PPD is a hair higher than two 2-cores.

The 2 quads still running 2.5.3 (both Q6600s) are doing two 2-core VMs. One of these days I'll get around to updating them.