Hyper-V Guest not performing as expected

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

katit

Active Member
Mar 18, 2015
436
31
28
53
Ok, I'm currently working in VM(Windows 8) on my MacBook Pro 13. I use VMWare Fusion. My guest OS Win8. I do .NET development, Visual Studio 2012 and XAML..

VM works OK, but we all want more, right? Since I built new server (dual E5-2670, 128Gb RAM - from Deals section) - I was toying with an idea of moving my dev VM into Hyper-V and work using remote desktop.
This would be beneficial in many ways:

1. I won't loose VM if I loose laptop.
2. Hyper-V replica will be great for backing up
3. VM will be on same network with databases (faster when I work from home)
4. I don't have to buy most expensive MacBook with tons of RAM and big SSD...

So.. for VM storage on server

I installed 2x 400Gb Intel 3710 SSDs and created Stripe (in Windows server)
I converted disk to VMDX
VM didn't launch with SCSI, had to put it on EDI
Setup VM with 32Gb of RAM (have 8 in Mac)
-16 CPUs (have 3 in Mac)
-pass through GPU from Host.

Everything runs! Great. I have nice GUI in remote desktop and appears like all is well.
But. When I run my not-so-scientific test of compiling large .NET solution ..

Takes 60 seconds on Mac, 90 seconds in Hyper-V!

What is the problem? It is much slower in Visual studio launching projects, programs open slower..

Attached is disk speeds measured inside VM on Mac and on new server.
Also, on Mac I have 2.9Ghz CPU vs 2.6 on server

Anything else I should look at? I'm really upset as it's not useable as it is right now :(
 

Attachments

gigatexal

I'm here to learn
Nov 25, 2012
2,913
607
113
Portland, Oregon
alexandarnarayan.com
Depending on your model of mac the pcie ssd could be quicker than the setup you have in Windows. Try benching just one disk in Windows and compare.

Also how exactly did you pass through your GPU to the VM on hyper V? I thought that didn't work all that well.
 

katit

Active Member
Mar 18, 2015
436
31
28
53
What you see attached is storage benchmarked INSIDE VM on both Mac and "Server"
Sure it is faster on Mac, it's 2015 model with 512Gb so IT IS fast. And it shows on those tests even virtualized.

On Server I have 2 of the Intel 3710 (from deals section) STRIPED - so it's not very bad but worse than one under Fusion.

But do you think this is it? I really don't want to throw money at PCIe SSD now :)

GPU: I passed it using Hyper-V RemoteFX.
With or without GPU - this specific test on compiling solution is about same.

With GPU passthrough I get much better UI on RDP. It feels good, almost as good as when I work on local VM under Fusion. So it definitely helps for visual aspect of it. But not sure if it's needed for what I use this VM.
 
  • Like
Reactions: gigatexal

katit

Active Member
Mar 18, 2015
436
31
28
53
Did some more tests...

Let's call VMs "HyperV" and "Fusion" to make it easier..

HyperV:
Compiles in 90 sec
Removed GPU pass-through. Didn't affect anything. 90 sec.
Installed RAM disk (16Gb). Copied all solution there. Speeds like 3x of what's in Fusion.. Time went down to 80 seconds. Not much at all.
Noticed windows "Malware" scanner activity. Disabled live scanning. Compile on RAM disk - 60 seconds!

- Shut down and reduced cores to 3 like in Fusion - 80 seconds.


Fusion:
Compiles in 60 seconds
Removed malware scanner - 45 seconds


So far conclusion is not conclusive :( Seems like CPU frequency plays role as well? 2.9 on Mac vs 2.59 in Xeon? Lot's of cores help (because msbuild parallels those tasks), but not as much as high frequency?
 

katit

Active Member
Mar 18, 2015
436
31
28
53
Nope. It's Bradwell. Only 1 CPU/2 cores. VMWare sees 4 cores (hyperthreading) and I give 3 to my VM
But 2670 boosts to 3.3 according to Intel..

See comparison from Intel. 2.9 is boosted speed on laptop CPU. There should be no question who wins here.. It's just weird.. Can't be so much difference in hypervisor??
 

Attachments

vBuild2BAU

Member
Feb 27, 2016
41
9
8
Melbourne Australia
Nope. It's Bradwell. Only 1 CPU/2 cores. VMWare sees 4 cores (hyperthreading) and I give 3 to my VM
But 2670 boosts to 3.3 according to Intel..

See comparison from Intel. 2.9 is boosted speed on laptop CPU. There should be no question who wins here.. It's just weird.. Can't be so much difference in hypervisor??
First of all on your Hyper-V VM try 8 vCPU instead.

Second the other part is data locality. How close the data is to the compute and memory.
Which is why the RAM Drive is makes a big difference.
Explained really well by Chad Sakac here :
(skip to 40mins 30secs)

Third of all did a VM should ideally have less than the number of physical cores backing it but never more than the number of physical cores. Hyperthreaded cores don't count. So your Fusion VM should be at a MAX of 2 cores.

Hope that helps.
 

gigatexal

I'm here to learn
Nov 25, 2012
2,913
607
113
Portland, Oregon
alexandarnarayan.com
Nope. It's Bradwell. Only 1 CPU/2 cores. VMWare sees 4 cores (hyperthreading) and I give 3 to my VM
But 2670 boosts to 3.3 according to Intel..

See comparison from Intel. 2.9 is boosted speed on laptop CPU. There should be no question who wins here.. It's just weird.. Can't be so much difference in hypervisor??
Single core boost to 3.3 yeah
 

katit

Active Member
Mar 18, 2015
436
31
28
53
Also how exactly did you pass through your GPU to the VM on hyper V? I thought that didn't work all that well.
Btw, I'm removing this card. I didn't see any performance difference in what I do. But there was couple other issues, licensing, BIOS, drivers. And at the end - Chrome or FireFox would blink and display artefacts. I guess RemoteFX is not quite ready.
 

katit

Active Member
Mar 18, 2015
436
31
28
53
First of all on your Hyper-V VM try 8 vCPU instead.

Second the other part is data locality. How close the data is to the compute and memory.
Which is why the RAM Drive is makes a big difference.
I will try 8.

Data is on SATA3 channels on physical hosts. 2 Intel 3710 SSDs in STRIPE. Stripe made with Windows Server 2012 Host. You can see those speeds is OK.

With RAM Drive - speeds ON TEST were much better (like 3x of what I see under Fusion) but I got only 10% of improvement in compile time and it is still much worse than under Fusion. So, incredible disk speed helps but not radical and still slower.
 

katit

Active Member
Mar 18, 2015
436
31
28
53
It's Windows 8, only IE10 :(
Yes, it works fine.

P.S. How do I upgrade to 8.1? Store doesn't have option to download anymore and I can't find any other definite instructions
 

keybored

Active Member
May 28, 2016
280
66
28
Make sure you have all the Windows 8 updates installed. My understanding is that the Store won't show the option to upgrade if you're behind on updates. It's also entirely possible that MS removed the option altogether expecting you to upgrade to Win10 instead...
 

katit

Active Member
Mar 18, 2015
436
31
28
53
Make sure you have all the Windows 8 updates installed. My understanding is that the Store won't show the option to upgrade if you're behind on updates. It's also entirely possible that MS removed the option altogether expecting you to upgrade to Win10 instead...
All updates installed. It's a dev VM, there is a tons of config and setup involved, probably 2 days of setting up. I was postponing upgrade and now I'm stuck with it..

I tried Windows 10 in VM and it seems even slower :(