Re: [ovirt-users] Network performance drop when compared to other hypervisor with vhost_net on for UDP
Check files under /etc/tune-profiles/ HTH Martin Pavlik RHEV QE > On 14 Jan 2015, at 13:55, mad Engineer wrote: > > Thanks Martin, > How can we see the changes made by tuned? > for "virtual guest" i see it changes scheduler to deadline.Is there > any way to see what parameters each profile is going to change > > Thanks > > On Wed, Jan 14, 2015 at 6:14 PM, Martin Pavlík wrote: >> Hi, >> >> from the top of my head you could try to play with tuned both with guest and >> host >> >> ###Install### >> yum install tuned >> /etc/init.d/tuned start >> chkconfig tuned on >> >> ###usage### >> list the profile: >> tuned-adm list >> >> change your profile: >> tuned-adm profile throughput-performance >> >> maybe try to experiment with other profiles. >> >> HTH >> >> Martin Pavlik >> RHEV QE >> >>> On 14 Jan 2015, at 12:06, mad Engineer wrote: >>> >>> I am running RHEL6.5 as Host and Guest on HP server. >>> Server has 128G and 48 Core[with HT enabled.] >>> >>> 3 VMs are running 2 pinned to first 24 PCPU with proper NUMA pinning, >>> >>> Guests: >>> >>> VM1: >>> 6 VCPU pinned to 6 PCPU NUMA node 1,with 16G RAM >>> >>> VM2: >>> 6 VCPU pinned to 6 PCPU on NUMA node 0,with 16G RAM >>> >>> VM3: >>> 2 VCPU ,no pinning,4G RAM >>> >>> HOST >>> host has 10 free CPU+24 HT threads which is not allocated and is available. >>> Host also runs a small application that is single threaded,that uses ~4G >>> RAM. >>> >>> Total resource to host is 10 CPU+24 HT=34 and 92G unallocated RAM[VMS >>> dont even use 70% of allocated RAM] also ksm is not running. >>> >>> Networking: >>> Uses linux bridge connected to 1Gbps eth0,with ip assigned on eth0 >>> [This IP is called for accessing application running on host] >>> All vms use virtio and VHOST is on . >>> >>> Traffic on virtual machines are ~3MBps and combined traffic on host is >>> ~14MBps >>> >>> "VHOST-pid-of-qemu-process" sometimes uses ~35% CPU. >>> >>> >>> There is no packet loss,drop or latency,but the issue is with the same >>> setup on Vmware with same sizing of virtual machines,with the only >>> difference as application running on host has moved to fourth VM.So in >>> Vmware there are 4 VMs. >>> Application gives better number ie on KVM that number is 310 and on >>> vmware it is 570.Application uses UDP to communicate. >>> >>> I tried removing VHOST,still value is same.(I hope VHOST-NET UDP issue >>> is solved) >>> >>> Thanks for any help >>> ___ >>> Users mailing list >>> Users@ovirt.org >>> http://lists.ovirt.org/mailman/listinfo/users >> ___ Users mailing list Users@ovirt.org http://lists.ovirt.org/mailman/listinfo/users
Re: [ovirt-users] Network performance drop when compared to other hypervisor with vhost_net on for UDP
Thanks Martin, How can we see the changes made by tuned? for "virtual guest" i see it changes scheduler to deadline.Is there any way to see what parameters each profile is going to change Thanks On Wed, Jan 14, 2015 at 6:14 PM, Martin Pavlík wrote: > Hi, > > from the top of my head you could try to play with tuned both with guest and > host > > ###Install### > yum install tuned > /etc/init.d/tuned start > chkconfig tuned on > > ###usage### > list the profile: > tuned-adm list > > change your profile: > tuned-adm profile throughput-performance > > maybe try to experiment with other profiles. > > HTH > > Martin Pavlik > RHEV QE > >> On 14 Jan 2015, at 12:06, mad Engineer wrote: >> >> I am running RHEL6.5 as Host and Guest on HP server. >> Server has 128G and 48 Core[with HT enabled.] >> >> 3 VMs are running 2 pinned to first 24 PCPU with proper NUMA pinning, >> >> Guests: >> >> VM1: >> 6 VCPU pinned to 6 PCPU NUMA node 1,with 16G RAM >> >> VM2: >> 6 VCPU pinned to 6 PCPU on NUMA node 0,with 16G RAM >> >> VM3: >> 2 VCPU ,no pinning,4G RAM >> >> HOST >> host has 10 free CPU+24 HT threads which is not allocated and is available. >> Host also runs a small application that is single threaded,that uses ~4G RAM. >> >> Total resource to host is 10 CPU+24 HT=34 and 92G unallocated RAM[VMS >> dont even use 70% of allocated RAM] also ksm is not running. >> >> Networking: >> Uses linux bridge connected to 1Gbps eth0,with ip assigned on eth0 >> [This IP is called for accessing application running on host] >> All vms use virtio and VHOST is on . >> >> Traffic on virtual machines are ~3MBps and combined traffic on host is >> ~14MBps >> >> "VHOST-pid-of-qemu-process" sometimes uses ~35% CPU. >> >> >> There is no packet loss,drop or latency,but the issue is with the same >> setup on Vmware with same sizing of virtual machines,with the only >> difference as application running on host has moved to fourth VM.So in >> Vmware there are 4 VMs. >> Application gives better number ie on KVM that number is 310 and on >> vmware it is 570.Application uses UDP to communicate. >> >> I tried removing VHOST,still value is same.(I hope VHOST-NET UDP issue >> is solved) >> >> Thanks for any help >> ___ >> Users mailing list >> Users@ovirt.org >> http://lists.ovirt.org/mailman/listinfo/users > ___ Users mailing list Users@ovirt.org http://lists.ovirt.org/mailman/listinfo/users
Re: [ovirt-users] Network performance drop when compared to other hypervisor with vhost_net on for UDP
Hi, from the top of my head you could try to play with tuned both with guest and host ###Install### yum install tuned /etc/init.d/tuned start chkconfig tuned on ###usage### list the profile: tuned-adm list change your profile: tuned-adm profile throughput-performance maybe try to experiment with other profiles. HTH Martin Pavlik RHEV QE > On 14 Jan 2015, at 12:06, mad Engineer wrote: > > I am running RHEL6.5 as Host and Guest on HP server. > Server has 128G and 48 Core[with HT enabled.] > > 3 VMs are running 2 pinned to first 24 PCPU with proper NUMA pinning, > > Guests: > > VM1: > 6 VCPU pinned to 6 PCPU NUMA node 1,with 16G RAM > > VM2: > 6 VCPU pinned to 6 PCPU on NUMA node 0,with 16G RAM > > VM3: > 2 VCPU ,no pinning,4G RAM > > HOST > host has 10 free CPU+24 HT threads which is not allocated and is available. > Host also runs a small application that is single threaded,that uses ~4G RAM. > > Total resource to host is 10 CPU+24 HT=34 and 92G unallocated RAM[VMS > dont even use 70% of allocated RAM] also ksm is not running. > > Networking: > Uses linux bridge connected to 1Gbps eth0,with ip assigned on eth0 > [This IP is called for accessing application running on host] > All vms use virtio and VHOST is on . > > Traffic on virtual machines are ~3MBps and combined traffic on host is ~14MBps > > "VHOST-pid-of-qemu-process" sometimes uses ~35% CPU. > > > There is no packet loss,drop or latency,but the issue is with the same > setup on Vmware with same sizing of virtual machines,with the only > difference as application running on host has moved to fourth VM.So in > Vmware there are 4 VMs. > Application gives better number ie on KVM that number is 310 and on > vmware it is 570.Application uses UDP to communicate. > > I tried removing VHOST,still value is same.(I hope VHOST-NET UDP issue > is solved) > > Thanks for any help > ___ > Users mailing list > Users@ovirt.org > http://lists.ovirt.org/mailman/listinfo/users ___ Users mailing list Users@ovirt.org http://lists.ovirt.org/mailman/listinfo/users
[ovirt-users] Network performance drop when compared to other hypervisor with vhost_net on for UDP
I am running RHEL6.5 as Host and Guest on HP server. Server has 128G and 48 Core[with HT enabled.] 3 VMs are running 2 pinned to first 24 PCPU with proper NUMA pinning, Guests: VM1: 6 VCPU pinned to 6 PCPU NUMA node 1,with 16G RAM VM2: 6 VCPU pinned to 6 PCPU on NUMA node 0,with 16G RAM VM3: 2 VCPU ,no pinning,4G RAM HOST host has 10 free CPU+24 HT threads which is not allocated and is available. Host also runs a small application that is single threaded,that uses ~4G RAM. Total resource to host is 10 CPU+24 HT=34 and 92G unallocated RAM[VMS dont even use 70% of allocated RAM] also ksm is not running. Networking: Uses linux bridge connected to 1Gbps eth0,with ip assigned on eth0 [This IP is called for accessing application running on host] All vms use virtio and VHOST is on . Traffic on virtual machines are ~3MBps and combined traffic on host is ~14MBps "VHOST-pid-of-qemu-process" sometimes uses ~35% CPU. There is no packet loss,drop or latency,but the issue is with the same setup on Vmware with same sizing of virtual machines,with the only difference as application running on host has moved to fourth VM.So in Vmware there are 4 VMs. Application gives better number ie on KVM that number is 310 and on vmware it is 570.Application uses UDP to communicate. I tried removing VHOST,still value is same.(I hope VHOST-NET UDP issue is solved) Thanks for any help ___ Users mailing list Users@ovirt.org http://lists.ovirt.org/mailman/listinfo/users