Re: [Condor-users] Condor installation: VMMEMORY setting behaves odd / buggy?

[Matthew Farrellee <matt@xxxxxxxxxx>]

On 08/11/2010 10:50 PM, Rob wrote:
> Hi,
>
> I'm about to install Condor on 300+ public library Windows PCs,
> which have VMware 1.0.10 installed.
> During the Condor installation I have to decide the value of VMMEMORY
> and VMMAXNUMBER
>
> > From the manual:
>
> VMMEMORY
> an integer value that defines the maximum memory each VM run on the
> target machine.
>
> VMMAXNUMBER
> an integer value that defines the number of VMs that can be run in parallel
> on the target machine.
> I want to run at most one VM per slot on a multi core PC, so I set
>     VMMAXNUMBER = NUMBER_OF_PROCESSORS
>
>
> I then decided:
>
> VMMEMORY = ( 50 * RAM ) / ( 100 * NUMBER_OF_PROCESSORS )
>
> or, in other words all the slots together share equally half of the PC's total
> RAM.
>
> (I thought that VMMEMORY should not be too large, as the Windows OS
> may still need some RAM, even though it is in idle state. Is that right?
> Or am I too conservative here?)
>
> For example: a dual core PC with 2GB of RAM will then assign 512 MB to each
> slot as VMMEMORY. I've tried this on a test PC with such specs, and I see:
>
> $ condor_status -vm
>
> Name VMType Network State Activity LoadAv VMMe ActvtyTime VMNetworking
>
> slot1@SP_1F_02  vmware nat,bridge Unclaimed Idle     0.930   512  0+00:00:04
> slot2@SP_1F_02  vmware nat,bridge Unclaimed Idle     0.050    12  0+00:08:10
>                  Total Owner Claimed Unclaimed Matched Preempting Backfill
>
>    INTEL/WINNT51     2     0       0         2       0          0        0
>            Total     2     0       0         2       0          0        0
>
> $ condor_status -vm -l
>
> MyType = "Machine"
> TargetType = "Job"
> VM_Type = "vmware"
> VM_Memory = 512
> VM_Networking = TRUE
> VM_Networking_Types = "nat,bridge"
> VM_AvailNum = 10000
>
> MyType = "Machine"
> TargetType = "Job"
> VM_Type = "vmware"
> VM_Memory = 512
> VM_Networking = TRUE
> VM_Networking_Types = "nat,bridge"
> VM_AvailNum = 10000
>
>
> So it seems that each slot on this PC can use 512 MB RAM.
>
> ===================================================
>
> Now, when a virtual machine with "vm_memory = 500" runs on
> one slot, I will get following:
>
> $ condor_status -vm
>
> Name VMType Network State Activity LoadAv VMMe ActvtyTime VMNetworking
>
> slot1@SP_1F_02  vmware nat,bridge Claimed   Busy  0.000  12  0+00:00:34
> slot2@SP_1F_02  vmware nat,bridge Unclaimed Idle   0.010  12  0+00:08:57
>                  Total Owner Claimed Unclaimed Matched Preempting Backfill
>
>    INTEL/WINNT51     2     0       1         1       0          0        0
>            Total     2     0       1         1       0          0        0
>
> $ condor_status -vm -l
>
> MyType = "Machine"
> TargetType = "Job"
> VM_Type = "vmware"
> VM_Memory = 12
> VM_Networking = TRUE
> VM_Networking_Types = "nat,bridge"
> VM_AvailNum = 10000
>
> MyType = "Machine"
> TargetType = "Job"
> VM_Type = "vmware"
> VM_Memory = 12
> VM_Networking = TRUE
> VM_Networking_Types = "nat,bridge"
> VM_AvailNum = 10000
>
>
> This is what confuses me:
> the VM_Memory has reduced to 12 MB on BOTH slots!!
>
> This means: the 512 MB assigned to each slot, is actually the TOTAL
> memory the two slots can use TOGETHER.
> This seems to be in conflict with the description in the manual of VMMEMORY,
> which says that EACH virtual machine (= each slot?) can use the
> specified memory.....
>
> Is something buggy here?
>
> Thank you!
> Rob.

The doc is buggy here.

According to the 7.4 manual,

VM_MEMORY
    An integer to specify the maximum amount of memory in Mbytes that 
will be allowed to the virtual machine program.

VM_MAX_NUMBER
    An integer limit on the number of executing virtual machines. When 
not defined, the default value is the same NUM_CPUS. When it evaluates 
to Undefined, as is the case when not defined with a numeric value, no 
meaningful limit is imposed.

The VM_MEMORY description is still a bit vague. It would be more clear 
as "will be allocated to virtual machines." Both these params define a 
startd-wide resource (memory and # vms) to be consumed by running VMs. 
So VM_MEMORY is really the max shared among all VMs.

https://condor-wiki.cs.wisc.edu/index.cgi/tktview?tn=625,31

It may be interesting to generalize these resources. You might say we 
have pool wide limiters in concurrency limits, but no host local limits, 
except for these.

Best,


matt