40ed108dd5
KVM-based microVMM for the Volt platform: - Sub-second VM boot times - Minimal memory footprint - Landlock LSM + seccomp security - Virtio device support - Custom kernel management Copyright (c) Armored Gates LLC. All rights reserved. Licensed under AGPSL v5.0
3.1 KiB
3.1 KiB
Volt Network Benchmark Results
Test Environment
| Parameter | Value |
|---|---|
| Date | YYYY-MM-DD |
| Host CPU | Intel Xeon E-2288G @ 3.70GHz |
| Host RAM | 64GB DDR4-2666 |
| Host NIC | Intel X710 10GbE |
| Host Kernel | 6.1.0-xx-amd64 |
| VM vCPUs | 4 |
| VM RAM | 8GB |
| Guest Kernel | 6.1.0-xx-amd64 |
| QEMU Version | 8.x.x |
Test Configuration
- Duration: 30 seconds per test
- Ping count: 1000 packets
- iperf3 parallel streams: 8 (multi-stream tests)
Results
Throughput (Gbps)
| Test | virtio | vhost-net | macvtap |
|---|---|---|---|
| TCP Single Stream | |||
| TCP Multi-8 Stream | |||
| UDP Maximum | |||
| TCP Reverse |
Latency (microseconds)
| Metric | virtio | vhost-net | macvtap |
|---|---|---|---|
| ICMP P50 | |||
| ICMP P95 | |||
| ICMP P99 | |||
| TCP P50 | |||
| TCP P99 |
Packets Per Second
| Packet Size | virtio | vhost-net | macvtap |
|---|---|---|---|
| 64 bytes | |||
| 128 bytes | |||
| 256 bytes | |||
| 512 bytes |
Transaction Rates (trans/sec)
| Test | virtio | vhost-net | macvtap |
|---|---|---|---|
| TCP_RR | |||
| TCP_CRR | |||
| UDP_RR |
Analysis
Throughput Analysis
TCP Single Stream:
- virtio: X Gbps (baseline)
- vhost-net: X Gbps (Y% improvement)
- macvtap: X Gbps (Y% improvement)
Key Finding: [Describe the performance differences]
Latency Analysis
P99 Latency:
- virtio: X µs
- vhost-net: X µs
- macvtap: X µs
Jitter (P99/P50 ratio):
- virtio: X.Xx
- vhost-net: X.Xx
- macvtap: X.Xx
Key Finding: [Describe latency characteristics]
PPS Analysis
64-byte Packets (best overhead indicator):
- virtio: X pps
- vhost-net: X pps (Y% improvement)
- macvtap: X pps (Y% improvement)
Key Finding: [Describe per-packet overhead differences]
Conclusions
Performance Hierarchy
-
macvtap - Best for:
- Maximum throughput requirements
- Lowest latency needs
- When host NIC can be dedicated
-
vhost-net - Best for:
- Multi-tenant environments
- Good balance of performance and flexibility
- Standard production workloads
-
virtio - Best for:
- Development/testing
- Maximum portability
- When performance is not critical
Recommendations
For Volt production VMs:
- Default:
vhost-net(best balance) - High-performance option:
macvtap(when applicable) - Compatibility fallback:
virtio
Anomalies or Issues
[Document any unexpected results, test failures, or areas needing investigation]
Raw Data
Full test results available in:
results/virtio/TIMESTAMP/results/vhost-net/TIMESTAMP/results/macvtap/TIMESTAMP/
Reproducibility
To reproduce these results:
# On server VM
iperf3 -s -D
sockperf sr --daemonize
netserver
# On client VM (for each backend)
./run-all.sh <server-ip> virtio
./run-all.sh <server-ip> vhost-net
./run-all.sh <server-ip> macvtap
# Generate comparison
./compare.sh results/
Report generated by Volt Benchmark Suite