evga LinSMP Vs. NotFred

[jfb9301]

Is there an advantage to those of us to run evga's LinSMP distro compaired to the NotFred Distro?

I'm not looking for -bigadv.

I'm looking for advantages for my memory limited quad core.

 

[Zero82z]

There shouldn't be any advantage. And in what way are you having memory issues? notfred's only needs 1GB at most.

 

[jfb9301]

not enough ram to even give -bigadv a shot.

I keep having:

Attempt to access beyond end of device
hda: rw=1, want=2097008, limit=2088387

or line after line after line of that with the want number going slowly upappear on my notfred for my Q9650....it's flipping annoying.

I was hoping the LinSMP distro might not do that and might have some advantages as notfred hasn't been updated in ages.

I'm allocating 1024MB to the VM BTW

 

[Zero82z]

I don't think that's even a memory issue. But if you want to get rid of it, just go ahead and install a full Linux distro. That's what I did, and I haven't had any problems with it since I made the VM over a year ago.

 

[jfb9301]

there is that, as an option, I was thinking doing a full install there would be some loss of folding power running all the stuff that I did not turn off. Tho I could install ubuntu, install folding, set up folding to run with run level 3 and set the machine to only boot to RL3, skipping the GUI to save process power.

that would also mean digging up a copy of VM-ware workstation or getting a machine made for me.

 

[T4rd]

not enough ram to even give -bigadv a shot.

I keep having:

Attempt to access beyond end of device
hda: rw=1, want=2097008, limit=2088387

or line after line after line of that with the want number going slowly upappear on my notfred for my Q9650....it's flipping annoying.

I was hoping the LinSMP distro might not do that and might have some advantages as notfred hasn't been updated in ages.

I'm allocating 1024MB to the VM BTW

All of my 4 difference PCs give me that same message in Notfreds, so I don't think it's a memory issue or anything either. I have no idea what causes it, but it doesn't affect its performance, so I don't mind it, heh.

 

[APOLLO]

not enough ram to even give -bigadv a shot.

I keep having:

Attempt to access beyond end of device
hda: rw=1, want=2097008, limit=2088387

or line after line after line of that with the want number going slowly upappear on my notfred for my Q9650....it's flipping annoying.

I was hoping the LinSMP distro might not do that and might have some advantages as notfred hasn't been updated in ages.

I'm allocating 1024MB to the VM BTW

I agree with Zero, that it's likely another issue. I never seen those messages before. I have allocated as little as 2.5GB of RAM to my VMs for testing -bigadv. I even reduced the threads to 7 in order to determine minimum required resource allocation to run these units. What happened in my case is the client stalled without any CPU activity. I have left it for over 30 minutes and no CPU activity ever commenced within the experimentation period. I just gave up on it.

You will likely require ~4GB of memory or around that much. I have only 4GB in a native installation that works but cannot verify if that is truly the minimum in native or VM environments. It is a worthwhile endeavor to test and discover true rock-bottom minimum hardware requirements for -bigadv, IMHO.

 

[Zero82z]

there is that, as an option, I was thinking doing a full install there would be some loss of folding power running all the stuff that I did not turn off.

No performance difference for me between running on a full Fedora 10 distro and running notfred's. As for making the VM, you can do that with the Workstation trial, or you can use Server which is free (although it only supports 2-core VMs, but you just edit the .vmx file once you've made it to solve that problem).

 

[SmokeRngs]

Currently I have a native openSUSE 11.1 installation on my server. The workload on the machine is rather minimal for the most part. I have it running an XP VM for certain duties but it's not CPU intensive at all. The system is also my fileserver as well as my secondary machine for browsing/chatting if my main machine is otherwise occupied. It's a Q6600@3.6 and puts out about 6200PPD running the Linux client natively.

Currently, I have a cut down "full" install of openSUSE 11.2 running in a VM on my main system (Q6600@3.6) alongside the GPU client within Win7. It's currently doing around 5600PPD when I'm not actively using the machine. The PPD numbers on this VM is a bit lower than I was expecting. Previously I was able to pull 200-400 PPD more with this machine (when not in use) while running 2 VMs with 2 cores. Those two older VMs were running cut down "full" installs of openSUSE 10.3. All of these VMs were running KDE.

That all said, I've never actually used notfreds or any other premade VM so I'm not sure what the performance would be like on them. However, I would not expect them to do much, if any better. Considering the VM just sits there minimized and runs the F@H client, the GUI should not be using much, if any CPU resources which is going to be the main factor regarding folding.

Outside of not wanting to setup a distro on your own, the main reason to use a prebuilt Linux VM for folding is to minimize the hard drive space for the VM and to a smaller extent the amount of memory usage for the VM. My openSUSE 11.2 VM is currently using 526 meg of RAM out of an allocated 768 meg. I'm thinking about dropping it down to 512 meg because the extra RAM is not needed and the most likely things to get swapped will be the unused OS/GUI components. This should still leave the SMP client with all the RAM it needs to avoid any slowdowns.

 

[APOLLO]

Currently, I have a cut down "full" install of openSUSE 11.2 running in a VM on my main system (Q6600@3.6) alongside the GPU client within Win7. It's currently doing around 5600PPD when I'm not actively using the machine. The PPD numbers on this VM is a bit lower than I was expecting. Previously I was able to pull 200-400 PPD more with this machine (when not in use) while running 2 VMs with 2 cores. Those two older VMs were running cut down "full" installs of openSUSE 10.3. All of these VMs were running KDE.

That's an interesting performance result. Many people here have posted that converting dual-core VMs to one single VM results in superior performance. I have always doubted it and in my case it's still up in the air whether it's better in regards to PPD. I will have to look at old logs in my shut down VMs and compare with the sole running VMs to extrapolate the results.

One thing for sure, the systems are definitely a lot snappier and far easier to get running from a reboot than having as many as 4 VMs running concurrently. Perhaps more importantly, Stanford prefers the much quicker turn around times and I will leave things they way they are for these reasons.

 

[Zero82z]

That's an interesting performance result. Many people here have posted that converting dual-core VMs to one single VM results in superior performance. I have always doubted it and in my case it's still up in the air whether it's better in regards to PPD.

I saw a definite increase after the 2.10 core was released, in any case. I went from 2000-2200PPD per VM to 4800-5800PPD for the single VM (depending on if/how the system is being used for other things).

 

[APOLLO]

I saw a definite increase after the 2.10 core was released, in any case. I went from 2000-2200PPD per VM to 4800-5800PPD for the single VM (depending on if/how the system is being used for other things).

I think on my systems it's a mixed bag but cannot confirm it at this point because I have to restart the old VMs and look into the logs. For example, one system is averaging 6.5 minutes per frame now, and before with 4 VMs, each one was averaging about 20-22 minutes. However, on another system with identical motherboard but different processors: 4 VMs were averaging about 17-18 minutes per frame, now does around 3.5-4 minutes per frame. This is from memory without checking the actual logs.

One question: are you using the "export MPICH_NO_LOCAL=1" command?

 

[SmokeRngs]

That's an interesting performance result. Many people here have posted that converting dual-core VMs to one single VM results in superior performance. I have always doubted it and in my case it's still up in the air whether it's better in regards to PPD. I will have to look at old logs in my shut down VMs and compare with the sole running VMs to extrapolate the results.

One thing for sure, the systems are definitely a lot snappier and far easier to get running from a reboot than having as many as 4 VMs running concurrently. Perhaps more importantly, Stanford prefers the much quicker turn around times and I will leave things they way they are for these reasons.

I'm somewhat confused with the result with the VM myself. I was expecting to see similar performance in the VM to my native Linux install. The only reason I created this new VM with 11.2 was because my previous VMs with 10.3 were running terribly slow even after making the changes so it ran on 4 cores instead of 2. For some reason, I was only getting about 4500PPD at the best of times. As I said, I'm doing 6200PPD with the single SMP client on my native Linux system and before I managed similar numbers between that native install and dual VMs.

 

[The_Heretic]

All of my 4 difference PCs give me that same message in Notfreds, so I don't think it's a memory issue or anything either. I have no idea what causes it, but it doesn't affect its performance, so I don't mind it, heh.

I see it on one of the NotFreds. That's a C2Q with 8 gigs of ram running Win7 64. It's not affecting the box's performance though.

 

[zero2dash]

Yeah I see that error on both of my VM Notfred boxes (Q9450 + E8400). Nothing bad happens so I ignore it. I searched this (DC) forum yesterday and saw other people saying they had the error with no problems and they ignore it also.... I think it's a disk write error or something but it's not messing with the client and not causing it to crash so whatever.

 

[Zero82z]

One question: are you using the "export MPICH_NO_LOCAL=1" command?

No. I had tried it but didn't see any difference.

 

[Mr. Pedantic]

I like the eVGA Linux distro a lot better than Notfred's. It's a lot more friendly and intuitive to set up, and it also allows using extra flags - the one thing I was most annoyed about with Notfred was that the log file wasn't as verbose as I would have liked it to be. Performance seems about similar, so I think the eVGA Linux distro is better for me.

No. I had tried it but didn't see any difference.

Can you use the MPICH_NO_LOCAL command with the eVGA LinSMP?

 

[Zero82z]

Can you use the MPICH_NO_LOCAL command with the eVGA LinSMP?

I have no idea; I've never used the EVGA distro.