ZStack Interpretation | After VMware’s shift to a subscription model, how can users choose virtualiz

Recently, the 2024 China Cloud Computing Infrastructure Developers Conference (hereinafter referred to as the CID Conference) was held in Beijing. This conference showcased the cutting-edge technological innovations in the field of cloud computing infrastructure, bringing together numerous technical experts and industry pioneers to discuss the latest developments and future trends in cloud computing infrastructure. ZStack, as a leading infrastructure software provider, was invited to attend the event, where senior development engineer Wang Boce shared ZStack’s VF network card live migration technology.

In network device virtualization technology, although Virtio network cards can meet most virtualization needs, they are prone to latency and packet loss issues under high CPU loads. In contrast, SR-IOV VF network cards, with their low latency and low CPU utilization, have become the preferred choice for high-performance applications. However, the hot migration of SR-IOV VF network cards has been limited by insufficient support in QEMU. Although the industry has proposed net-failover solutions and migration strategies for OpenStack, these solutions either depend on specific environments or result in excessively long business downtime.

ZStack’s VF network card hot migration technology offers a new solution by balancing compatibility and performance. This technology covers pure virtualized network cards, Virtio network cards, and hardware-virtualized SR-IOV VF network cards, the latter of which is preferred for high-performance network cards due to its low latency, low CPU resource usage, and high PPS (packets per second).

Compared to other industry solutions, ZStack’s technical solution has clear advantages. Compared with the Macvtap solution, ZStack’s solution performs better; compared with the net-failover solution, ZStack’s solution supports a broader range of kernel versions; compared with hardware vendor solutions, ZStack’s solution offers better compatibility; compared with the OpenStack solution, ZStack’s solution results in shorter traffic interruption times during hot migration.

ZStack’s hot migration technology includes preparations before migration, unloading network cards and routing switching during migration, and mounting new VF network cards and restoring network configurations after migration. This process ensures that virtual machines use Virtio network cards during hot migration and VF network cards at other times, allowing for quick switching and significantly reducing network downtime.

The focus of ZStack’s technology is the rapid switching between VF and Virtio. The core of ZStack’s proposed solution is how to quickly switch between these two modes to minimize network downtime as much as possible. ZStack has implemented the following optimizations:

• Unified MAC Address: VF and Virtio network cards use the same MAC address, which prevents network devices from needing to relearn the virtual machine’s ARP table and Layer 2 forwarding table, thus reducing network downtime.
• Routing-based Traffic Switching: Compared to link aggregation, using routing changes to switch network traffic is more stable and reliable, especially avoiding potential traffic interruption issues in link aggregation on older operating systems.
• Optimized Configuration Sequence: By using GuestTools to configure the network and reasonably arranging the configuration sequence of VF and Virtio network cards, it further reduces the delay caused by internal traffic switching in the virtual machine.

ZStack has effectively solved these challenges through its innovative solution. During hot migration, ZStack uses Virtio network cards temporarily to replace VF network cards, avoiding dependence on specific QEMU versions or GuestOS and significantly reducing business downtime to the second level. This method not only maintains the high performance of SR-IOV VF network cards but also enables fast and reliable hot migration, providing a balanced solution that combines compatibility and performance for network virtualization in cloud computing.