Performance Tuning of the CSR 1000V
To improve performance of a guest VM in a hypervisor environment, you improve availability of the I/O and other hardware resources through para-virtualization. Para-virtualization allows for a kernel to present a software interface to a guest VM that is similar but not identical to that of the underlying hardware, thereby improving the VM performance. If you want to tune the performance further, you need to look at two components:
Hypervisor scheduling
CPU pining
This section reviews the common tuning practices for an ESXi host. The scheduler for ESXi is responsible for vCPU, IRQ (interrupt requests), and I/O threads. To provide equal treatment to all guest VMs, the scheduler provides allocation of equal resources of vCPU threads for scheduling. Note that you can relax coscheduling of threads to avoid synchronization latency.
To tweak the scheduling and resource allocation details, you must access the VM setting using vSphere client and follow these steps:
In the vSphere client inventory, right-click the virtual machine and select Edit Settings.
Click the Resources tab and select CPU.
Allocate the CPU capacity for this virtual machine.
The Processor Affinity setting (CPU pining) restricts VMs to a particular set of cores by defining the affinity set. The scheduling algorithm aligns with process affinity for assigning the resources used for the tasks. Figure 4-49 assumes two tasks: Task 1 and Task 2. Task 1 has affinity to processor 1 and is using it. When Task 2 needs a resource, the scheduler uses a second processor. Task 2 then acquires affinity with the second processor.
)
Figure 4-49 CPU Pining
To tweak these settings, access the vSphere client and follow these steps:
In the vSphere client inventory panel, select a virtual machine and select Edit Settings.
Select the Resources tab and select Advanced CPU.
Click the Run on Processor(s) button.
You achieve CPU pining in KVM by issuing the following command:
sudo virsh vcpupin test 0 6
Hyperthreading by definition allows a single physical core to have two logical cores; that is, a single core can execute two threads at a given time. Each process from the guest VM can be split into multiple threads to a logical CPU, and the CPU can handle multiple threads of independent tasks. The main function of hyperthreading is to increase the number of tasks in the pipeline by creating parallel pipelines. By tweaking the process affinity option, you can restrict VMs to a particular set of cores and unhook the VM from processor scheduling. Most of the hypervisors use BIOS settings to modify the hyperthreading feature.
For predictable performance, the following best practices are recommended:
Ensure that hyperthreading is turned off.
Use CPU pining to allow the guest VMs to dedicate one or more physical hardware CPUs for processing.
For CSR 1000V performance optimization, it is important to understand the concept of DirectPath I/O and SR-IOV (single root I/O virtualization). These are driver virtualization and are beneficial for achieving very high packet rates with low latency. In DirectPath I/O, you can map only one physical function to one virtual machine. SR-IOV allows an admin to share a single physical device, so that multiple virtual machines can connect directly to the physical function.
These features are supported in all hypervisors, and it is important to understand the settings on the hypervisor deployed in order to optimize guest VM performance with features used on the hypervisor.