February 20, 2023
OpenShift and Performance
Edge computing has become crucial in the realm of Telco/5G as it plays a significant role in addressing latency and congestion issues, as well as enhancing application performance. In simpler terms, the speed at which data packets move from the sender to the receiver after processing is determined by latency.
To fulfill the network performance criteria of 5G, it is imperative to maintain a network architecture with the least possible latency delay. Compared to 4G technology, which has an average latency of 50 ms, the goal of 5G is to achieve latency figures of 1 ms or less, resulting in a ten-fold increase in wireless throughput. In the Telco industry, numerous deployed applications demand minimal latency and cannot endure any packet loss. Adjusting a system for zero packet loss is vital in mitigating latency problems that degrade network performance.
Edge computing helps to reduce the overall latency by moving computing closer to end users, resulting in reduced application response times and performance latency.
The OpenShift Container Platform currently offers techniques to optimize software for low latency (with a response time of approximately <20 microseconds) and real-time execution on an OpenShift Container Platform cluster. These techniques involve kernel tuning, adjusting OpenShift Platform settings, installing a new real-time kernel, and reconfiguring the machine.
Fortunately, all of these steps can now be executed by the Node Tuning Operator for automated platform tuning. This performance profile configuration simplifies the process of making changes more reliably for the cluster administrator. The administrator can specify settings such as updating the kernel to kernel-rt, reserving CPUs for the cluster and operating system housekeeping tasks (including pod infra containers), and isolating CPUs for running workloads in application containers.
The OpenShift and Metalvisor combination provides bare-metal-like performance by leveraging Metalvisor's TypeZero hypervisor operating under the operating system. The Metalvisor hypervisor is launched from the firmware UEFI, providing a trusted and secure foundation for virtualized workloads. This architecture enables Metalvisor to provide new levels of determinism and Quality of Service (QoS) for workloads, just like bare-metal, but with the added benefits of virtualization.
The Metalvisor hypervisor provides hardware-level virtualization, which offers improved performance and resource management, allowing for more efficient use of system resources and better performance for workloads. By providing a more secure, reliable, and performant virtualization layer, Metalvisor andOpenShift enables customers to run their workloads with confidence, even in demanding edge environments.
"Metalvisor dedicates hardware CPU, Cache, Memory, Storage, and Network to the VM resulting in proper workload sizing, determinism, and guaranteed quality of service (QoS). "
Metalvisor does not share hardware between VMs; it statically maps resources to VMs and provides the max utilization of those resources to the VMs. This gives guest VMs the same profile as bare-metal (and also how we got the name Metalvisor).
Metalvisor, as a TypeZero hypervisor, Has no orchestration layer at the virtual machine (VM) layer and has no oversubscription. This is very different from Type 1 & 2 hypervisors. By having no orchestration trying fairly share resources between VMs and not over-subscribing workloads, the virtualization tax/overhead is removed, and workloads can utilize close to 100% of the hardware.
Metalvisor can even adjust how many cache ways that can be assigned to a VM giving it higher or lower determinism and hardware isolation. Metalvisor uses Cockpit for VM creation and Ansible for repeatable automation.
Metalvisor is able to guarantee the quality of service QoS that can result in more efficient and proper workload sizing. By having the guarantee of resources when you need them, architects can start to reduce the over-provisioning of workloads, such as additional vCPUs and Memory that has traditionally been added to compensate for the lack of resources and QoS that is lacking under traditional virtualization.
Now architects can properly right-size workloads and have confidence that workloads will have resources even when the system near capacity and workloads are are full committed.
The combination of Metalvisor and OpenShift can enhance the performance capabilities by running low latency, latency-sensitive and real-time workloads effectively due to its bare-metal-like performance, providing new levels of determinism and Quality of Service (QoS) for workloads.
Now architects may want to reassess particular views on virtualization and demanding workloads where they thought VMs would not be able to meet their performance requirements/demands. Metalvisor can help provide the bare-metal-like environment with hardware-based isolation and max utilization of hardware resources that are unmatched by type 1 & 2 hypervisors.
This combination of OpenShift and Metalvisor provides additional QoS and Determinism to run latency-sensitive workloads and real-time systems.