[libvirt-users] How to best I/O performance for Window2008 and MSSQL guest VM
Hi There, I've moved some Windows2012 with MSSQL VMs from an hold ESXi 5.5 machine to a more recent and powerful machine running Fedora 24 x86_64 and related libvirt + KVM virtualization. I've moved the VMs filesystem to LVM slices and installed the VirtIO drivers in to all Windows VMs. I've also set both Disk and Network interface to work using VirtIO. So far so good everything works pretty fine. Now I would like to tune at best the MSSQL VM for both Disk and Network interfaces in order to get the best performance possible. Regarding network interface I've set it to work as bridge instead to go through to macvtap, so I'm not sure what is the best in this case. Regarding disk, since it's LVM I've chosen to go to cache mode none and IO mode native. Also here cannot judge what's the best setup for the workload. I'm undecided to use IO mode threads along directsync cache mode. Finally I've set "ionice -c 1 -p <qemu-pid> -n 0" and "renice -n -10 <qemu-pid>" for the interested VM that I want to get best performance possible. Even with the above setup, the MSSQL VM has performance similar to the old machine running ESXi 5.5, so does anyone can suggest where too look at and/or what would be the right setup? Thanks in advance. Roberto Fichera
Hi Roberto, What is the cpu and memory configuration of your guest? Are you pinning to dedicated CPUs, are you exposing host topology and cpu features, do you have dedicated I/O threads? Are you backing the guest memory with hugepages? All of the above will likely increase performance and minimise noise. Cheers, On 6 December 2016 at 05:13, Roberto Fichera <kernel@tekno-soft.it> wrote:
Hi There,
I've moved some Windows2012 with MSSQL VMs from an hold ESXi 5.5 machine to a more recent and powerful machine running Fedora 24 x86_64 and related libvirt + KVM virtualization. I've moved the VMs filesystem to LVM slices and installed the VirtIO drivers in to all Windows VMs. I've also set both Disk and Network interface to work using VirtIO. So far so good everything works pretty fine. Now I would like to tune at best the MSSQL VM for both Disk and Network interfaces in order to get the best performance possible.
Regarding network interface I've set it to work as bridge instead to go through to macvtap, so I'm not sure what is the best in this case.
Regarding disk, since it's LVM I've chosen to go to cache mode none and IO mode native. Also here cannot judge what's the best setup for the workload. I'm undecided to use IO mode threads along directsync cache mode.
Finally I've set "ionice -c 1 -p <qemu-pid> -n 0" and "renice -n -10 <qemu-pid>" for the interested VM that I want to get best performance possible.
Even with the above setup, the MSSQL VM has performance similar to the old machine running ESXi 5.5, so does anyone can suggest where too look at and/or what would be the right setup?
Thanks in advance. Roberto Fichera
_______________________________________________ libvirt-users mailing list libvirt-users@redhat.com https://www.redhat.com/mailman/listinfo/libvirt-users
-- Cheers, ~Blairo
On 12/06/2016 06:06 AM, Blair Bethwaite wrote:
Hi Roberto,
Hi Blair
What is the cpu and memory configuration of your guest?
I've set to copy host configuration (16 cores) and memory is set to 24GB, host has 64GB. Guest is Windows 2012 64bits version Architecture: x86_64 CPU op-mode(s): 32-bit, 64-bit Byte Order: Little Endian CPU(s): 16 On-line CPU(s) list: 0-15 Thread(s) per core: 2 Core(s) per socket: 8 Socket(s): 1 NUMA node(s): 1 Vendor ID: GenuineIntel CPU family: 6 Model: 79 Model name: Intel(R) Xeon(R) CPU E5-2620 v4 @ 2.10GHz Stepping: 1 CPU MHz: 1200.219 CPU max MHz: 3000.0000 CPU min MHz: 1200.0000 BogoMIPS: 4194.79 Virtualization: VT-x L1d cache: 32K L1i cache: 32K L2 cache: 256K L3 cache: 20480K NUMA node0 CPU(s): 0-15 Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc aperfmperf eagerfpu pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch epb intel_pt tpr_shadow vnmi flexpriority ept vpid fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm cqm rdseed adx smap xsaveopt cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local dtherm ida arat pln pts
Are you pinning to dedicated CPUs,
no
are you exposing host topology and cpu features,
Yes
do you have dedicated I/O threads?
if you are talking about disk I/O, no. Should I do that?
Are you backing the guest memory with hugepages?
Don't know, how can I check or set it?
All of the above will likely increase performance and minimise noise.
Cheers,
On 6 December 2016 at 05:13, Roberto Fichera <kernel@tekno-soft.it> wrote:
Hi There,
I've moved some Windows2012 with MSSQL VMs from an hold ESXi 5.5 machine to a more recent and powerful machine running Fedora 24 x86_64 and related libvirt + KVM virtualization. I've moved the VMs filesystem to LVM slices and installed the VirtIO drivers in to all Windows VMs. I've also set both Disk and Network interface to work using VirtIO. So far so good everything works pretty fine. Now I would like to tune at best the MSSQL VM for both Disk and Network interfaces in order to get the best performance possible.
Regarding network interface I've set it to work as bridge instead to go through to macvtap, so I'm not sure what is the best in this case.
Regarding disk, since it's LVM I've chosen to go to cache mode none and IO mode native. Also here cannot judge what's the best setup for the workload. I'm undecided to use IO mode threads along directsync cache mode.
Finally I've set "ionice -c 1 -p <qemu-pid> -n 0" and "renice -n -10 <qemu-pid>" for the interested VM that I want to get best performance possible.
Even with the above setup, the MSSQL VM has performance similar to the old machine running ESXi 5.5, so does anyone can suggest where too look at and/or what would be the right setup?
Thanks in advance. Roberto Fichera
_______________________________________________ libvirt-users mailing list libvirt-users@redhat.com https://www.redhat.com/mailman/listinfo/libvirt-users
Hi Roberto, On 6 December 2016 at 19:01, Roberto Fichera <kernel@tekno-soft.it> wrote:
I've set to copy host configuration (16 cores) and memory is set to 24GB, host has 64GB. Guest is Windows 2012 64bits version
Is the guest configured with 16 cpus or 16 cores?
Are you pinning to dedicated CPUs,
no
I would, to ensure cpu or cache performance is absolutely not a problem, and it's simple to do (see vcpupin).
are you exposing host topology and cpu features,
Yes
So you have something like: <cpu mode='host-passthrough'> <topology sockets='1' cores='8' threads='2'/> In your domain xml?
do you have dedicated I/O threads?
if you are talking about disk I/O, no. Should I do that?
Hard to say what the problem is without a lot more information - maybe the system in question is performance limited by the DB design... But if you are bottlenecking on I/O in the guest whilst the host seems to have more I/O available then you probably want to consider tuning this: https://libvirt.org/formatdomain.html#elementsIOThreadsAllocation
Are you backing the guest memory with hugepages?
Don't know, how can I check or set it?
You can either use transparent huge pages (assuming you are not doing any PCI passthrough), or create a static hugepage allocation for guests. Search for documentation specific to your OS. -- Cheers, ~Blairo
participants (2)
-
Blair Bethwaite -
Roberto Fichera