6:22 p.m.
Hi All
We are currently testing our product using KVM as the hypervisor. We are
not using KVM as a bare-metal hypervisor. We use it on top of a RHEL
installation. So basically RHEL acts as our host and using KVM we deploy
guests on this system.
We have all along tested and shipped our application image for VMware
ESXi installations , So this it the first time we are trying our
application image on a KVM hypervisor.
On this front i have done some tests to find out how our application's
response time is when deployed on KVM and then compare it with a VM
deployed on VMware ESXi. We have a benchmark test that basically loads
the application simulating a load of 100 parallel users logging into the
system and downloading reports. These tests basically use a HTTP GET
query to load the application VM. In addition to that i have taken care
to use the same hardware for both the tests , one with RHEL(Host)+KVM
and another with VMware ESXi. All the hardware specifications for both
the servers remain the same. The load test also remains the same for
testing with both the servers.
First observation is that the average response time on the VMware ESXi
is : 500 milli-seconds while the application's average response time
when deployed using RHEL(Host)+ KVM is : 1050 milli-seconds. The
response time of the application when deployed on KVM is twice as much
as when it is deployed using VMware ESXi.
I did few more tests to find which sub-system on these servers shows
varying metrics.
First i started with IOZone to find out if there is any mismatch in the
speed with which data is read / written to the local disk on the two VMs
and found that "Read" speed in the VM that was deployed using
RHEL(Host)+KVM was twice as slow as the VM which was deployed using
VMware ESXi.
For more on IoZone , Please refer : http://www.iozone.org/
more specifically the following IoZone metrics were twice as less when
compared to the server running with VMware ESXi:
Read
Re-read
Reverse-Read
Stride Read
Pread
Note: I had run the IoZone tests on the VMs on both the servers.
Second observation to be made was the output from the "top" command. I
could see that the VM deployed on RHEL(Host)+KVM was showing high
numbers for the following metrics when compared with the VM deployed on
VMware ESXi:
load averages
%sy for all the logical processors
%si for all the logical processors
i debugged further to find out which device is causing more interrupts
and found it to be "ide0" , See the output from the /proc/interrupts
file below:
The other interrupts apart from ide0 are pretty much similar to the VM
deployed using VMware ESXi.
************/proc/interrupts *******************
[root@localhost ~]# cat /proc/interrupts
CPU0 CPU1 CPU2 CPU3 CPU4 CPU5
CPU6 CPU7
0: 795827 0 0 0 0 0
0 0 IO-APIC-edge timer
1: 65 0 0 0 0 0
0 0 IO-APIC-edge i8042
6: 2 0 0 0 0 0
0 0 IO-APIC-edge floppy
8: 0 0 0 0 0 0
0 0 IO-APIC-edge rtc
9: 0 0 0 0 0 0
0 0 IO-APIC-level acpi
10: 425785 0 0 0 0 0
0 0 IO-APIC-level virtio0, eth0
11: 47 0 0 0 0 0
0 0 IO-APIC-level uhci_hcd:usb1, HDA Intel
12: 730 0 0 0 0 0
0 0 IO-APIC-edge i8042
14: 188086 0 0 0 0 0
0 0 IO-APIC-edge ide0
NMI: 0 0 0 0 0 0
0 0
LOC: 795813 795798 795783 795767 795752 795737
795723 795709
ERR: 0
MIS: 0
*********************************************
Any pointers to improving the response time for the VM for
RHEL(Host)+KVM installation would be greatly appreciated.
Thanks
Jatin