Red Hat Bugzilla – Bug 1335309
Set optimal clock source in Windows guests using the guest agent
Last modified: 2018-03-19 09:18:42 EDT
Description of problem: Customer has several storage domains using Fibre Channel to connect back to a Nimble storage array. We have applied specific recommended settings for Nimble already. We are testing using a VMWare Hypervisor/Host that connects to the same SAN switches on identical server hardware. We are not re-installing. There are 2 separate UCS blades with identical hardware setup. We are trying for a like-for-like configuration in every way. The data store in VMWare uses Fibre Channel to the Nimble the same as RHEV. The virtual disks are normal virtual disks. In the Testing Results docx file you'll see that the RHEV Windows 2012 R2 guest shows 11393 IO/sec and an average I/O response time of 7 msec, whereas the VMWare Windows 2012 R2 guest shows 65244 IO/sec and an average I/O response time of 1.2 msec. We would like to determine what can be done on the RHEV side to get better performance out of this VM. Version-Release number of selected component (if applicable): RHEV-M 3.5.7 Red Hat Enterprise Virtualization Hypervisor release 7.2 (20160105.1.el7ev) Windows 2012 R2 Guests with virtio-win-1.7.2-2.el6.noarch How reproducible: 100% (see screenshots in the Testing Results doc) Steps to Reproduce: 1. Install Windows 2012 R2 as a guest in RHEV and VMWare 2. Install guest tools and virtio drivers in the guest 3. Use IOMeter to get performance results Actual results: VMWare performs approximately 5x better than RHEV on the same hardware Expected results: Similar performance results in both virtualization suites Additional info: Customer data coming in a future comment
Created attachment 1161220 [details] IOmeter 8k random wrts with SSD
Created attachment 1161221 [details] IOmeter 8k random wrts with Fibre
I/O performance of w2k12r2 guest on two types of I/O Fibre and SSD. Testing was performed on a w2k12r2 guest with updates. Testing was done using IOmeter with a 20GB ntfs formated disk partion to exceed the guest memory. Different cache modes and aio were used. 1. Hardware - white box 2 Intel(R) Xeon(R) CPU E5-2699 v4 @ 2.20GHz - 256GB RAM - Non-Volatile memory controller: Intel Corporation PCIe Data Center SSD 2. Hardware - white box 4 Intel(R) Xeon(R) CPU X7560 @ 2.27GHz - 128GB RAM - 2 ISP2532-based 8Gb Fibre Channel, 1 HP HSV300 3. Guest machine: - Win2012r2 w/updates - virtio-win-0.1.117-1 drivers - 1 and 7 vCPU - 10GB RAM - 30GB raw provisioned disk - 50GB virtio disk
Moving to agent, based on https://bugzilla.redhat.com/show_bug.cgi?id=1277353#c28
Same fix and impact as: https://bugzilla.redhat.com/show_bug.cgi?id=1354532 Copied information: Windows XP, Windows Vista and Windows 2003 Server are no longer supported by us and therefore ignored Windows 2008 Server does not have the USEPLATFORMCLOCK option at all and therefore it's not implemented - From Windows 2008 R2 and above and from Windows 7 and above we're now deleting the value from the configuration - Unless the ovirt guest agent configuration does not have set the 'general.apply_timer_configuration' value to false (if it is false we won't do anything)
Verified on 4.0.2-3
Since the problem described in this bug report should be resolved in a recent advisory, it has been closed with a resolution of ERRATA. For information on the advisory, and where to find the updated files, follow the link below. If the solution does not work for you, open a new bug report. https://rhn.redhat.com/errata/RHEA-2016-1700.html