• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

Issues with Linux VM's and SMP Client

L

lelliott731

[H]ard DCOTM April 2017
Joined
Mar 15, 2002
Messages
941
So I finally set up my dual Xeon box, and I downloaded VMWare Server v1-0-6. Downloaded Ubuntu v8-04 and set up 4 copies of Ubuntu each with two CPU cores. They all seem to be doing pretty well, except when I monitor the CPU usage on the Windows Server 2003 Enterprise Edition x64 operating system, I see a good bit of wasted CPU cycles and I'm not sure if I'm doing something wrong or not.

Here is an image of what the task manager sees:

CPU_Utilization.jpg


As you can see, 6 of the eight cores are doing 90% or more CPU work, but two of them have much less then that and I'm just curious if anyone else has had issues like this.

Thanks!
 
my dual VMs tend to hover around 85-90 per core, but I've never seen single cores just drop out like that. you already have the smp clients running on these, correct?

 
Yeah, that's what is so weird. I have dual Quad Core Xeon's, no hyperthreading or anything like that, and I have 4 instances of Ubuntu 8.04 running each with two processors selected in the VMWare Server instance, it's very odd.
 
Ok, so after messing around and setting the affinity for each of the VMWare Server VM's, this is now what it looks like:

CPU_Utilization2.jpg


CPU0 & CPU1 are for the first VM, CPU2 & CPU3 are for the second VM & so on, so it looks like they are not making full use of each of the two cores on the VM instance.

Does anyone know why that might be happening?
 
It's pretty normal. Even with 2 VMWare instances on a quad, it is possible there is still free cycles wasted.

As a example to maximize it fully, my Q9450 have 2 VMWare instances running SMP each, 1 GPU2 instance with a 8800GTS 640MB and 1 uniprocessor instance. I get about 8500 ppd total.

 
Damn dude, you run 2 SMP clients on 2 processors??? Or do you let them just take whatever spare CPU cycles they can find?

The way I have my Q6600 setup is that I have my GPU2 client on one core, my uniprocesor client on one core, and an SMP client on the other two.
 
That looks normal to me.
I find that with the VM, when you set affinity, one core runs at 100% and the other is slightly down from 100%.
How much down depends on the protien.
But my total with 4 VM's is normally around 90-93%

One trick.
Set affinity to run the 4 VM's on 7 cores as this does not slow them down much.
You now have an empty core to run a GPU client on .......... :p

Luck ........... :D
 
Ok, thanks Tigerbiten, now if only the server boards came with PCIe x16 slots, all I have is PCIe x8 slots...stupid server boards!!!
 
lelliott731 said:
Ok, thanks Tigerbiten, now if only the server boards came with PCIe x16 slots, all I have is PCIe x8 slots...stupid server boards!!!
Click to expand...

I'm lucky.
My X7DAL-E has a PCIE x16 slot.
Its the slower of my two Xeon MoBo by around 400 PpD.
But add 5,000 PpD from the GPU client will more than make up the difference.
Downside is that I cannot use my Thermalright Xeon coolers if theres a full lenght vid card in.
So it back to the passive coolers with 92mm fans on top.

Edit. Skulltrails have 4 x16 slots ...... :p

Luck ........... :D
 
Tigerbiten said:
I'm lucky.
My X7DAL-E has a PCIE x16 slot.
Its the slower of my two Xeon MoBo by around 400 PpD.
But add 5,000 PpD from the GPU client will more than make up the difference.
Downside is that I cannot use my Thermalright Xeon coolers if theres a full lenght vid card in.
So it back to the passive coolers with 92mm fans on top.

Edit. Skulltrails have 4 x16 slots ...... :p

Luck ........... :D
Click to expand...
Thanks for the info Tigerbiten, I went with the 5100 chipset just because I didn't need the speed of the FB-DIMM's I didn't think, and I wanted to save energy & heat, although now that the GPU2 clients are available, I think I've made a mistake.
 
Woohoo...I just found a MicroATX 5100 Motherboard, and it has a PCIe x16 Slot!!!
It's a SuperMicro X7DCA-L!!!
 
Yeah right brother, it'll be a few weeks before you'll have to start to worry, remember, I'm still about 10k ppd behind you, and my raise doesn't show up in my paycheck till July 15th.
 
So I think I finally got it fixed. I had to add a 5th instance of Ubuntu in a VMWare VM, and have it set up to run on the 2nd core of each of the other 4 Ubuntu instances. This is now what my CPU utilization looks like:

CPU_Utilization3.jpg
 
lelliott731 said:
So I think I finally got it fixed. I had to add a 5th instance of Ubuntu in a VMWare VM, and have it set up to run on the 2nd core of each of the other 4 Ubuntu instances. This is now what my CPU utilization looks like:
Click to expand...

If your running 5x VM's on 8 cores you will need to check your PpD.
Does the increase in CPU use outweight the decrease in the PpD of the clients.
Putting 4x VM's on 7 cores means that each VM runs ~100 PpD slower.
But thats ok when you add ~5,000 PpD from the GPU client runnning on its own core.

So is the slow down of your clients outweighted by the increase in PpD ?

Luck ............... :D
 
Yeah, that's the question I have Tigerbiten, the issue is, I've been having problems with getting FahMon to run on the VM's, and I was trying to set up Samba so that I could have one Windows copy of FahMon monitoring all the 5 VM's, but I keep getting a weird error, and so I haven't been able to monitor it.

Forum Post about Error
 
Tigerbiten said:
I'm lucky.
My X7DAL-E has a PCIE x16 slot.
Its the slower of my two Xeon MoBo by around 400 PpD.
Click to expand...
The x4 PCI-E slot on the X7DAL-E can still be used for a second video card, no? It's in a physical x16 slot so a video card should be compatible even if it runs slower. I've read of people actually making openings on x4 physical slots to get their video cards to fit and it works, but at reduced performance, so a x4/x16 slot should also work.

lelliott731 said:
Thanks for the info Tigerbiten, I went with the 5100 chipset just because I didn't need the speed of the FB-DIMM's I didn't think, and I wanted to save energy & heat, although now that the GPU2 clients are available, I think I've made a mistake.
Click to expand...
Why do you believe you made a mistake with the memory choice? Many users who run dual Xeon quads complain about the FB-DIMM memory as the single biggest downside on this platform. Even Skulltrail owners view that as the biggest disadvantage. They are expensive, run extremely hot especially at high bus speeds and don't really offer a significant performance advantage in most non-server situations. I don't think having that memory type will help with the GPU2 client, unless I'm totally mistaken on what is beneficial.

 
Well thanks for making me feel better about my purchase APOLLO. I have just been hearing about the throughput of the FB-DIMM's, so since I was stuck with only Dual Channel, I was thinking maybe that was causing some of my slowdown.
 
lelliott731 said:
Well thanks for making me feel better about my purchase APOLLO. I have just been hearing about the throughput of the FB-DIMM's, so since I was stuck with only Dual Channel, I was thinking maybe that was causing some of my slowdown.
Click to expand...
Well, I could be wrong about the performance if you're referring to the new 800MHz FB-DIMMs which support faster FSBs. My experience is with the 667MHz variety that I use on my systems. They run very hot at full load especially if the FSB is over 1333MHz, and you must have active cooling on the memory but that isn't always convenient with some setups. I can't even imagine how hot the modules must get on an OC Skulltrail, and it's one of the reasons I'm avoiding that board like the plague. I don't feel at all comfortable investing in additional FB-DIMM boards. In light of the fact that they are on the verge of obsolescence, it would not be a wise decision near the 2H of '08. So, I think you made a wise decision.

I have a question for you which is on topic, how much RAM do you have installed on your system and how much does each VM use? It looks like 4GB in the pics, but want to be sure. I have a couple of dual quads myself but have avoided VMs because I can't stand Linux. I made two attempts at Linux VM installation on my X7DAL-E and both failed. First time was with Notfreds and then Tiger's recommended CentOS. It's really funny, in the 90s I used to live on the DOS command line but over a decade of Windows GUI and one tends to get {s}oft... :(
 
Now for a strictly "dummy" question concerning SMP'ing with 2x LinSMP clients in Ubuntu v4.08 (Hardy Heron) I'm talkin' straight here (non VMW'ed clients), I realize this may be a VMWare directed thread, but I believed problems runin' two non VMW'ed clients was pretty freakin' close :confused:

1) how do you set cpu infinity in Linux, do you even have to set the chip core infinities like is very beneficial in windose because it allows you to use the "Affinity program".

1a) how the frig do you set two independant LinSMP clients up (divide the RAM in configuration, install the clients in their own folders, prey to tha' computer Gods or what :p)

2) does the cpu run much hotter runnin' two straight clients

3) is it true by runin' two straight LinSMP clients (non VMW'ed) that you don't have to worry about "trashing" one when you stop the other.

The main reason I'm asking is I've pretty much gone over to Linux for my 24/7 folding boxen and I would like to run two clients on my Q6600's without having to worry about a WU crapping out if I shut one side down :rolleyes:

Many, many thanks for any answers, comments or constructive criticism.

 
APOLLO, I only have 4GB of RAM on the system, I bought 8GB, but there is something wrong with the motherboard where it won't recognize the memory in the 3rd & 4th slots, but if I put it in the 5th & 6th, it freaks out. And with Xilikon's recent challenge, I can't afford to have my box out of comission, I need every PPD I can get!

VM's running on Windows is quite easy, especially with how simple Ubuntu is at setting up, I don't know if Ubuntu is the fastest version of Linux for Folding, but it works fine for me.

jws2346, I have no idea how to set affinity in Linux, because my VM's run on Windows, I just set it through the task manager on the host OS. And the same answer for all the rest of your questions, I really don't know, sorry I can't be of more help! I'm not that good at Linux, only recently started playing with it.
 
jws2346 said:
Now for a strictly "dummy" question concerning SMP'ing with 2x LinSMP clients in Ubuntu v4.08 (Hardy Heron) I'm talkin' straight here (non VMW'ed clients), I realize this may be a VMWare directed thread, but I believed problems runin' two non VMW'ed clients was pretty freakin' close :confused:

1) how do you set cpu infinity in Linux, do you even have to set the chip core infinities like is very beneficial in windose because it allows you to use the "Affinity program".

1a) how the frig do you set two independant LinSMP clients up (divide the RAM in configuration, install the clients in their own folders, prey to tha' computer Gods or what :p)

2) does the cpu run much hotter runnin' two straight clients

3) is it true by runin' two straight LinSMP clients (non VMW'ed) that you don't have to worry about "trashing" one when you stop the other.

The main reason I'm asking is I've pretty much gone over to Linux for my 24/7 folding boxen and I would like to run two clients on my Q6600's without having to worry about a WU crapping out if I shut one side down :rolleyes:

Many, many thanks for any answers, comments or constructive criticism.

Click to expand...

1. You don't have to worry about affinity in linux.

1a. Not really sure if it makes any difference, but I would set it for half of the memory in the box as a precaution against getting 2 memory hungry units that will use more than you have.

2. I dont have a quad, cant say. :(

3. I never say never, but it is not an issue on linux.


216
 
lelliott731 said:
Can you run a 64-bit Linux OS on top of the 32-bit version of Windows XP?
Click to expand...

All my Single Quads run 32 bit Windoze XP Pro -> VMWare Server -> 64 bit Linux -> Linux Client.
All my SDual Quads run 64 bit Windoze XP Pro -> VMWare Server -> 64 bit Linux -> Linux Client.

It make no difference to the OS inside the VM which OS is running outside the VM.
You just have to make sure the BIOS is set correctly, so you can run a 64 bit OS inside the VM.

Luck ............ :D
 
Tigerbiten said:
All my Single Quads run 32 bit Windoze XP Pro -> VMWare Server -> 64 bit Linux -> Linux Client.
All my SDual Quads run 64 bit Windoze XP Pro -> VMWare Server -> 64 bit Linux -> Linux Client.

It make no difference to the OS inside the VM which OS is running outside the VM.
You just have to make sure the BIOS is set correctly, so you can run a 64 bit OS inside the VM.

Luck ............ :D
Click to expand...

And processors with VT support, which exclude the E2xxx and E4xxx series (maybe E7xxx as well).

 
You must log in or register to reply here.
Back
Top