[ale] Lack of speed up

[DJPfulio at jdpfu.com]

I know nothing at all about Hyper-V. It does some odd things compared to all the other hypervisors I've used (which is most of them).

However, there are some truths I've learned over the decades.
* Never over allocate system resources.  If the server is swapping, you did it wrong.
* Always leave some RAM unallocated to VMs.  Used to be that needed to be 1GB, but I suspect it is 2-4GB with MS-Windows bloat now.  So, if the system has 64GB of RAM total, then no more than 60GB should be allocated to running VMs.
* Always leave at least 1 Core to the hostOS.  Managing VMs requires the host to do things.  If the system has 8 cores, then no more than 6-7 should be specified for guest VM use.  I always start with 1 vCore per VM and only increase it to 2 when absolutely needed, but I'm trying to run multiple VMs.
* Guests should use virtio drivers for networking and storage.
* Guest storage should be on SSDs, not HDDs.  If on HDDs, then fully pre-allocated, so it isn't resizing all the time.  I would say to use a volume manager for even better performance, but I don't know if that's true with MS-Windows or not.
* If the guest has a GUI, allocate enough fake GPU resources, but not too much. If it doesn't have a GUI, then 9MB ( the default under Linux VMs for vRAM ) is sufficient.

I've seen new, Core i7, computers brought to their knees over incorrect allocations of vCPU and vRAM with MS-Windows VM-hosts.

Ensure the host and guests are actually sharing the limited resources and that neither is a hog.

On 4/27/25 15:40, Leam Hall via Ale wrote:
> Hey all, I'm mildly stumped by a lack of performance issue, and could
> use insight.
> 
> A friend was running Hyper-V on a Windows laptop, and spun an Ubuntu
> image with 8 cores to run some Python based analysis on two datasets.
> The datasets were roughly 150 and 190 MB respectively, and running
> one command line with & took about three hours. All eight cores were
> roughly 45% utilized, though with Hyper-V I'm not sure if that's
> cores assigned to the Ubuntu instance, or on the host machine. There
> was no memory swapping, and top showed one Python process.
> 
> He wanted things a bit faster, so I suggested opening a second
> terminal and running the command on each dataset separately, not in
> series. The load on all 8 cores went up about 15%, there were two
> Python processes, but the processes still took 3 hours.
> 
> Is this a Hyper-V based issue, or something else?
> 
> Thanks!
> 
> Leam
>