Live USB folding environment (persistent)

[shbris]

anyone have any ideas or suggestions for a live folding environment, possibly ubuntu or some custom linux that someone might have helpfully made

running SMP client is essential, and GPU would be beneficial, and persistent changes would also be useful. i'd like this because i have sporadic access to various computers that i'm keen to use to help us hit that 100M, but they won't be able to run for long enough periods to individually finish WU on time

that make sense? any input would be appreciated

[watercooled]

I found this a while ago, bit too tired to read through and check it's the same one but I think it is, and it sounds like what you're after: http://reilly.homeip.net/folding/usb.html
Oh and there's more stuff here: http://reilly.homeip.net/folding/

[shbris]

that's great cheers!

would it work as well if i ran a live usb ubuntu, i remember trying it a while ago but linux n00b couldn't get smp to work for some reason. then might get gpu folding sorted as well

thoughts?

[watercooled]

In stock configuration, a Live Ubuntu USB will discard all changes and any saved files on a reboot. You'd have to properly mount the USB drive.
A proper install to a USB drive rather than a Live environment might be better, here is a great guide showing you how to do it on Debian: http://linux.koolsolutions.com/2009/...memory-drives/

[shbris]

cheers for all this. found some materials online, mostly for 9.04 ubuntu though. gonna spend an evening or two experimenting, because self contained USB bootable ubuntu set up with SMP and GPU folding would be sweet

[watercooled]

The software in the first link I posted is worth considering, while it doesn't support GPU it does do SMP and it comes with a nice web interface you can use to reconfigure stuff and check how folding is going remotely, etc. Since most systems won't have a GPU worth folding on (or capable of it), something like that would boot faster and be lighter on resources, not that folding needs much RAM anyway, mind.

[scaryjim]

I'd've thought the best way to go about it would be to virtualise a linux box on a USB stick and run it with a free virtualisation software. I believe QEMU can be run without installation, so should work from a USB stick (for Windows binaries go to the links page and look under "unofficial QEMU binaries". That way you don't have to mess around rebooting the target machines all the time: just stick in the USB stick and run your VM, then when you lose access to the machine shut down the folding client in the VM, turn off the VM making sure it saves the state, move to another PC and start up the VM again! OK, it won't be quite as quick as running natively, but it means you can run on any PC you fancy, at any time, always picking up from where you left off, without rebooting the computer!

[watercooled]

Oh that's just made me think of something else - if you do a proper install to a flash drive, there is a good chance it won't work on different computers, different computers require different drivers for example. So what scaryjim suggested would be a better option unless all the PCs you plan to run on are identical, and even then that way would be more convenient.
I still think folding USB stick is a good way to go for non-GPU stuff though.

Oh and I've just thought of a potential problem for a VM - how well does that handle GPU drivers?

Yet another option would be to just extract a non-install version of F@H to a USB stick and run with the -local flag.

[scaryjim]

Having downloaded QEMU and read through some of the documentation I think it would be pretty straightforward to set up a persistent folding virtual machine on a USB stick. It allows you to declare a specific a number of CPUs for an SMP system, so you could basically declare the VM to be a 2, 3 or 4 core machine and then run it on any hardware regardless of the available physical cores (although obviously if you tried to pretend a Pentium 4 was a quad core it'd probably fold quite slowly ). You can also pass through network connections to upload and download work units, although I seem to remember that last time I tried that (about 3 years!) I had difficulty making it work.

Still, all said it shouldn't be too hard to get a working folding VM on an 8GB or 16GB USB stick that would persist between computers and could be used on any system you fancied: worth having a play with, I would've thought

[Tattysnuc]

... I tried this a while back using Microsofts virtual machine with XP on it. The problem I had was not witht he CPU folding - that ran nearly as quick as runnning natively The problem was with the GPU driver - The MS virtualised environment has a "generic" driver, not an NVidia one. I don;t believe that Nvidia support virtualisation directly.

HOWEVER, after giving up, I did spot some articles about using Virual box and another one that do allow you to fold in.

[scaryjim]

There's a few VMs that have experimental GPU virtualisation, but not many. Personally I'd be tempted to run the GPUs natively and the CPU in a VM. It also occurred to me this morning that if you did that you could set the client affinity on the VM which would then apply to the folding inside the VM - so you wouldn't have the problem of the core restarting and unsetting all of your affinity settings.

After a bit more playing with QEMU (I managed to get it to boot off the Win 7 install iso virtualised inside XP on a Celeron laptop! Slow as anything, but I was still quite impressed ) it turns out the virtualising the network requires a TAP driver on windows that you can get as part of the openvpn package. I've also found a link to *just* the TAP driver for Windows x64 but it's a command line installer with minimal documentation However, I'm thinking that QEMU virtualised SMP client + native GPU clients might be the way to go when I have PCP finished...

[scaryjim]

Quick updated for anyone interested in a portable virtualised folding environment: I've been messing with Qemu again tonight and have installed a TAP driver under Windows XP and got it to bridge with my wireless card after a *lot* of messing around (apparently wireless cards don't play nicely with bridging in XP and you have to mess around with netsh ). However, I can't get the tap tp play with damn small linux, and the Win 7 install appears to stall after you click the "install Windows" button. I don't have an XP ISO handy at home some I'm going to burn one at work and try again either at the weekend or some time next week. However, given the amount you have to mess around with TAP and bridging network connections, it's not going to be good for churning through lots of WUs. However, if you've got access to lots of different computers for a few hours each then it would work to run through units a bit at a time, and you could upload and download at the end of each unit from a dedicated machine.

I'll continue my experiments and report back again

[.walls]

I did something similar to this a while back in my folding days (used to fold for another well known forum), but nothing as exotic as using qemu...

In response to comments above; linux doesn't really work like Windows - it isn't that tied to the hardware that you install it on - especially if you don't have an X server running (and why would you? you want all the cycles for folding, right?).

My exploits used Arch linux and involved me writing a fairly poor rc script that called screen (so that you can connect to the session and see the current output if you want), but you could get the same results from any distro - just don't install a DE or WM.

Unless of course it's now better to fold with virtual windows environments rather than the other way round... I have been away from this for a bit now!

[scaryjim]

OK, having spent almost all of Friday evening messing around with every imaginable network setting etc., I finally managed to install XP inside a QEMU VM (took for ever, presumably because I don't have the hardware accelerator downloaded yet!), I set up a network bridge with a TAP adapter, I debugged the fact that my wireless wouldn't connect as part of the bridge, I persuaded the XP VM to install drivers for the network card: but I could not, no matter how hard I tried, get the network requests from the VM to come through the TAP. It was only at that point that I read the documentation properly, and realised that if all you need is for the VM to be able to access the internet, you don't need to mess around with TAPs at all *doh*

So, QEMU user-mode networking basically creates a virtual router, advertises it to the VM on one of the private network IP segments, then sends any network requests from the VM out through your actual network adapters. And with a couple of parameter changes in my shortcut, sure enough I was booting up Win XP virtualised under Win XP (yeah, I know; right?) and accessing Google. So, with a successful network connection from the VM, and the VM full portable, there's no reason why I can't just stick that all on a USB stick and then run a preset folding environment on any computer I can sit down at. Seems pretty cool to me. Next thing will be to find the acceleration packages and see how much it boosts the speed (and also if they can be run prtably or if they need installing...).

[scaryjim]

right, more research and experiments later:

While QEMU can emulate SMP systems, it presents each CPU as a *physical* CPU: so standard desktop versions of Windows will only use 2 CPUs, regardless of how many you actually have.

Portable QEMU on its own gives pretty poor CPU performance. Emulating Windows in QEMU on my laptop (a 1.86GHz Core2-based Celeron) I downloaded and ran wprime. Natively a 32M conpletes in a bit less than 100s (which isn't too bad for a single core system). On the VM it takes 88s to perform 6% of the same calculation

So in short, I don't think QEMU is really a viable option for a portable folding setup But it was fun messing around with it