VM's and SMP

S

starsfan7

Limp Gawd
Joined
Apr 29, 2005
Messages
193
So I decided to switch over an SMP client to a VM and I seem to have fouled something up, although what I am not sure. I have just been playing around with a VM to see if I would get a tangible ppd boost from a VM over a winSMP client. The machine this is on is a Q9550 @3.31ghz with 2GB of RAM on an Asus p5e dx. OS is winxp-32. The problem I am having is that the client seems to not be downloading or using SMP WU's or core. If I was using a winSMP client, it's like the -smp argument isn't on the client.

It's also showing fahcore_78 as the core. That's not an a1 or an a2. Am I ignorant or did I mess something up? I am new to this virtualization stuff, be noobish...

(client does not show -smp argument in console, only -local and -forceasm. Is there a way to add arguments to the client in the VM?)
 
First off, the VM is going to need 1 gig of RAM to work. So you need to pick up another stick or two before you do any VM stuff. (unless you want to run your whole system on 1 gig of ram.)

Can you post the log file so we know what's going on in there.
 
If its a pure folding box then you'll probably just get away with 1000Mb VM size.

Is virtulization switched on in the bios ??
Without it the VM will not run a 64bit OS.

Did you use VMWare Player ??
If so did you set it to use 4 CPU's ??

How did you install the client ??

Luck ............ :D
 
1) Are you using VMWare? Virtualbox doesn't allow 64-bit guests to run off 32-bit hosts. And if you're running a Linux distro, then it has to be 64-bit to run the SMP client.
2) You really should have a bit more RAM. My VM uses about 1.3GiB of RAM minimum usually.
3) What distro is running? NotFred is the easiest to install and configure, but you can use any other distro as well.
 
Kendrak said:
First off, the VM is going to need 1 gig of RAM to work. So you need to pick up another stick or two before you do any VM stuff. (unless you want to run your whole system on 1 gig of ram.)

Can you post the log file so we know what's going on in there.
Click to expand...

The system is literally being used for nothing else, with the vm running my usage is only around 80% total, so I think it should be ok, but like I said, I am really more curious and just messing around with it.

I followed the install guide in the sticky, except I gave the vm.exe idle across all 4 cores instead of 2 (wasn't the guide for a dual?). The client is running just fine, it got to 40% or 50% on the WU it was on. The issue is not performance, like I said, it's not pulling down SMP cores or WU's. I am using VMware player, and everything else to the letter in the guide.

I'm not at the machine now, but I will put up a logfile when I can.

EDIT
I am going win7-64 in a few weeks, just waiting for my msdnaa approval. Do you have to run the 32 bit client in a 32 bit environment? I thought virtualization could run 64 bit in a 32 bit OS? New to all this...
 
starsfan7 said:
EDIT
I am going win7-64 in a few weeks, just waiting for my msdnaa approval. Do you have to run the 32 bit client in a 32 bit environment? I thought virtualization could run 64 bit in a 32 bit OS? New to all this...
Click to expand...
The host OS doesn't matter with VMWare. As long as you have hardware virtualization enabled, you can run a 64-bit guest from within a 32-bit host.
 
Why is everyone recommending a minimum of 1GB of memory for VMs when I have gotten away with as little as 460MB for well over a year? Most of my VMs had 512MB and I was running 14 VMs at one time. Unless your Linux distro is one of those super bloated editions with everything included, you can do with much less than 1GB.
 
APOLLO said:
Why is everyone recommending a minimum of 1GB of memory for VMs when I have gotten away with as little as 460MB for well over a year? Most of my VMs had 512MB and I was running 14 VMs at one time. Unless your Linux distro is one of those super bloated editions with everything included, you can do with much less than 1GB.
Click to expand...

I know when I had 640mb, I would get errors and my VM would reboot. Maybe one in five.
 
Kendrak said:
I know when I had 640mb, I would get errors and my VM would reboot. Maybe one in five.
Click to expand...
That never happened to me but I'm using a very stripped down version of Linux. However, this was when I only had two cores per VM. Now with VM Player 3.0, it's not the same ballpark anymore.

Question: How much more memory is required for each VM when we start configuring clients to utilize more cores than the traditional two? Does the memory footprint of the client vastly increase with higher thread counts?
 
Up to four cores the client will use the same amount of threads and therefore the same amount of RAM. I don't know about scaling to more than 4 cores. And of course -bigadv workunits require significantly more memory.
 
Zero82z said:
Up to four cores the client will use the same amount of threads and therefore the same amount of RAM. I don't know about scaling to more than 4 cores. And of course -bigadv workunits require significantly more memory.
Click to expand...
Yes, you're correct. I was so used to having two cores per SMP client all these years I completely overlooked the fact that the client will work with four threads whether there are cores for all of them or not. So, yes I'm inquiring about memory usage above four cores and regular SMP.
 
With regular work-units, your looking at ~220Mb per thread, so you need just over 900Mb for 4 threads if your not going to page the VM.
Hence the 1,000Mb min size.
My regular VM's are running at ~930Mb.
Double it for -SMP 8.

For -bigadv work-units, your looking at double the standard work-unit size per thread.
Min size is probably ~5Gb for an 8 thread VM.
My -bigadv VM with 8 threads running is using ~4,530Mb.
Double that again if your running 16 threads.

Luck .......... :D
 
So I just started over completely when I got back to the machine, fresh config of everything...

And now it works. *shrug*

Code: 
 --- Opening Log file [November 21 22:39:45] 


# SMP Client ##################################################################
###############################################################################

                       Folding@Home Client Version 6.02

                          http://folding.stanford.edu

###############################################################################
###############################################################################

Launch directory: /etc/folding/1
Executable: ./fah6
Arguments: -local -forceasm -smp 4 

Warning:
 By using the -forceasm flag, you are overriding
 safeguards in the program. If you did not intend to
 do this, please restart the program without -forceasm.
 If work units are not completing fully (and particularly
 if your machine is overclocked), then please discontinue
 use of the flag.

[22:39:45] - Ask before connecting: No
[22:39:45] - User name: starsfan7 (Team 33)
[22:39:45] - User ID not found locally
[22:39:45] + Requesting User ID from server
[22:39:45] - Machine ID: 1
[22:39:45] 
[22:39:45] Work directory not found. Creating...
[22:39:45] Could not open work queue, generating new queue...
[22:39:45] - Preparing to get new work unit...
[22:39:45] + Attempting to get work packet
[22:39:45] - Connecting to assignment server
[22:39:45] - Successful: assigned to (171.64.65.56).
[22:39:45] + News From Folding@Home: Welcome to Folding@Home
[22:39:46] Loaded queue successfully.
[22:40:02] + Closed connections
[22:40:02] 
[22:40:02] + Processing work unit
[22:40:02] Core required: FahCore_a2.exe
[22:40:02] Core not found.
[22:40:02] - Core is not present or corrupted.
[22:40:02] - Attempting to download new core...
[22:40:02] + Downloading new core: FahCore_a2.exe
[22:40:09] Verifying core Core_a2.fah...
[22:40:09] Signature is VALID
[22:40:09] 
[22:40:09] Trying to unzip core FahCore_a2.exe
[22:40:09] Decompressed FahCore_a2.exe (5509624 bytes) successfully
[22:40:09] + Core successfully engaged
[22:40:14] 
[22:40:14] + Processing work unit
[22:40:14] Core required: FahCore_a2.exe
[22:40:14] Core found.
[22:40:14] Working on Unit 01 [November 21 22:40:14]
[22:40:14] + Working ...
[22:40:14] 
[22:40:14] *------------------------------*
[22:40:14] Folding@Home Gromacs SMP Core
[22:40:14] Version 2.10 (Sun Aug 30 03:43:28 CEST 2009)
[22:40:14] 
[22:40:14] Preparing to commence simulation
[22:40:14] - Ensuring status. Please wait.
[22:40:15] Called DecompressByteArray: compressed_data_size=4834267 data_size=23979129, decompressed_data_size=23979129 diff=0
[22:40:15] - Digital signature verified
[22:40:15] 
[22:40:15] Project: 2669 (Run 15, Clone 168, Gen 104)
[22:40:15] 
[22:40:15] Assembly optimizations on if available.
[22:40:15] Entering M.D.
[22:40:22] Multi-core optimizations on
[22:40:25] .D.
[22:40:32] Multi-core optimizations on
[22:40:34]  (0%)
[22:46:01] Completed 2500 out of 250000 steps  (1%)
[22:51:25] Completed 5000 out of 250000 steps  (2%)
[22:56:54] Completed 7500 out of 250000 steps  (3%)
[23:02:17] Completed 10000 out of 250000 steps  (4%)
[23:07:44] Completed 12500 out of 250000 steps  (5%)
[23:12:55] Completed 15000 out of 250000 steps  (6%)
[23:18:09] Completed 17500 out of 250000 steps  (7%)
[23:23:34] Completed 20000 out of 250000 steps  (8%)
[23:29:00] Completed 22500 out of 250000 steps  (9%)

5157 ppd per fahspy vs 2200 with winSMP. I'll take it.
 
formatreinstall.jpg
 
Ok, now I can ask the questions I wanted to in the first place x|. I have the VM set to idle priority, and it is pushing all 4 cores on one instance. Now I have been lead to believe from what I have researched about using a VM is that it will play nicely with an NV gpu if the priority is set properly. How many NV gpu's could I feasibly get away with, with a VM running? How dependent will that be on system RAM and 32 vs 64 os etc. Obviously for ram and vram capacity, addressing, etc., 64 bit is the way to go, but in the interim, how many NV gpu clients could a Q9550 support with the VM running before I start hurting myself?

I have been ati and winsmp until 2 weeks ago, all this is new to me.

Also, this machine is running the VM, one instance of pidgin, and a weather widget. Not much else.
 
You will max out your mobo, before you take up even close to one core with the needs of a bunch of GPUs.
 
starsfan7 said:
how many NV gpu clients could a Q9550 support with the VM running before I start hurting myself ?
Click to expand...

All my Q6600's @3.1 Ghz run 1x 4core VM set to idle and 3x nVidia GPU clients set to normal with no loss of points.

The max number of nVidia clients I have run on one box was 8, 4x 9800GX2's.
Which is the max possible due to driver issues.

Luck ........... :D
 
I have three GPU clients running via WINE in my secondary box which is also running a native Linux SMP client. I see a small PPD hit from running three clients compared to the one I was previously running, but I suspect a portion of that is due to WINE and not the clients themselves. It's also a dual-core CPU and not a quad.
 
You must log in or register to reply here.
Back
Top