Slow Disk Performance 2016 Standard VM vs 2008R2 VM RRS feed

  • Question

  • Let me preface this by starting off thanking anyone who can shed any light on this, as I've been banging my head off the wall for over a week now.

    The hardware in question:

    HP ProLiant ML150 Gen 9

    2 x Intel Xeon E5-2620v4 CPU

    32gb ram

    Host array:

    2 x 600gb sas 10k 2.5" drives in Raid 1

    Data Array:

    6 x 600gb sas 10k 2.5: drive in raid 1+0

    Array Controller:

    HP H240 Smart Host Bus Adapter

    Now before I get started, I realize now that the raid card does not have an accelerator, nor battery, and is thus not a full featured high performance card. However, this server was the highest spec that we could introduce within the current budget without going to the ML350 at a cost that would have put us over budget.

    Now the backstory...

    I converted two 2008r2 standard server deployments using disk2vhd and created vhdx disk images that I then loaded on the new server. These are set to static memory, 4 vcpus and dynamically expanding disks. These are both gen 1 deployments.

    I then created a new 2016 standard vhdx fixed disk, with 20gb of static memory, and 16 vcpus on a gen 2 deployment.

    On both the 2008r2 vm's i'm getting nearly 80mb/s over a 1gb nic connection to our backup device. I'm happy with this performance. However, on the 2016 vm, I can only manage about 25 mb/s to the same devices. 

    The server has 4 NICs in it, two Intel, and two Broadcom. I believe that the add in card is the Intel, with two ports, and the integrated is Broadcom with two ports. I've changed the nic in the VM for this server from Broadcom to Intel with no change in the performance. Local data installs show the same slower disk performance - so it also doesn't appear to be NIC related - although I'd be happy to look into that as well - as I've had issues with older drivers and VMQ before, but all firmware/driver updates have been deployed since this is a brand new box.

    I will provide some disk perf reports in the morning, as I have a new base image being taken for each vm by our backup currently.

    I've tried both a dynamic disk and a fixed disk for the 2016 server as well, and saw a small performance gain from that. I also changed from dynamic memory to static, but didn't see any performance changes there. I've ran out of ideas at this point - any feedback is greatly appreciated.

    Thursday, June 1, 2017 4:08 AM

All replies