Instances tested
The test results below are for the following.
OVH Elite VPS 2023
- Offered on 01/03/2023
- 8 Cores, 16GB of memory and 320GB of storage on the node os-bhs2
- Intel Core (Haswell, no TSX) (8) @ 2.399996GHz
- kvm on OpenStack Nova 19.3.2
- Network Interface: Virtualized, no speed provided due to virtio-pci driver.
OVH Elite VPS 2020 (Not tested, just for comparison)
- Offered on 01/01/2020
- 4 Cores, 8GB of memory and 160G of storage on the node os-bhs6
- Intel Core (Haswell, no TSX) (4) @ 2.399996GHz
- kvm on OpenStack Nova 14.1.1
- Network Interface: Virtualized, no speed provided due to virtio-pci driver.
OVH Dedicated Server Advanced LE
- Offered on 01/03/2022
- AMD Ryzen 5 5600X – 6c/12t – 3.7 GHz/4.6 GHz processor, 16GB memory, and 2x960GB NVMe disks
- ASRockRack X470D4U2-2T motherboard.
- Network Interface (ethtool enp1s0f0 | grep -i speed): Speed: 10000Mb/s
- Public bandwidth: 1 Gbps
- Private bandwidth: 1 Gbps unmetered
OVH Dedicated Server Advanced LE Virtual Machine Instance
- Proxmox VM Guest using KVM
- 4 cores, 8GB of memory and 120GB of storage.
Testing Notes
Getting Interface Link Speed
You want to know what the interface link speed of your instance is; this is possible with dedicated servers/baremetal. But not with virtualized guest/instances due to the use of a virtual network interface.
For dedicated servers/bare metal you can run ethtool to get the network interface speed. You’ll of course need the interface name
❯ ethtool enp1s0f0 | grep -i speed Speed: 10000Mb/s
Speed Tests
Higher speeds than network connection link speed with iperf3
You might notice that the OVH VPS iperf3 test is above (2000Mbps) than the operating system’s actual connection (1000Mbps). Trying to chase this down with iperf3, disabling all kinds of optimizations on the OS.
The likely culprit is that the instance/guest is deployed onto a node/host with a single 10G or multiple 1G connections. Essentially meaning there is more network capacity than 1000Mbps available.
The virtio-pci driver simply dumps network requests onto the bus to be dealt with, and there is no physical device on the instance/guest; therefore, there is no physical limit until the network traffic hits the physical network device. You can read more about it here. There is a good explanation of this on a ServerFault thread.
https://serverfault.com/questions/738840/get-link-speed-of-an-virtio-net-network-adapter
However, when looking at the interface used on both the OVH VPS and OVH Dedicated server using iftop or cmb both are interface bandwidth monitoring tools. Neither interface had surpassed the 1Gb/s interface limit.
Ookla Linux CLI versus speedtest-cli
There were two tests you could use via the CLI for speedtest.net.
- speedtest-cli from Github – https://github.com/sivel/speedtest-cli
- Ookla linux cli – https://www.speedtest.net/apps/cli
I don’t know the difference; they both provide similar functionality.
Ookla speedtest.net Inaccurate
As the title says, the Ookla speedtest.net test is inaccurate; there were multiple instances where the OVH Dedicated Server could get an above 1Gb/s download speed, sometimes 9Gb/s, as you can see below.
I was incorrect in my findings with Ookla speedtest.net being inaccurate; why? Google Search, when entering in 9177Mbps to GB/s, Google doesn’t check the case of letters. It sees 9177mbps to gb/s, which would mean “bits” and not bytes.
As you can see below, these tests are accurate.
9177.18 Mbits/s (bits) = 1147 MBytes/s (bytes) = 8.9620 Gbit/s (bits) = 1.1202 GBytes/s (bytes)
9413.97 Mbits/s (bits) = 1176 MBytes/s (bytes) = 9.1933 Gbit/s (bits) = 1.1491 GBytes/s (bytes)
Here’s another example, but from the command line, the Mbps is bits, and GB is bits.
As you can see, if you change the units of the speed test from the default to -u GB/s it doesn’t add up to the 9382.65 Mbps shown with the default units. If you do the conversion 9382.65 Mbps = 9.38265 GB/s, which is nine times the speed as the test using the GB/s units. I reached out to Ookla support.
At this point, I’m going crazy because I thought I knew how to convert speeds. But now I’m second-guessing myself.
I could have figured this out by just doing the math, and thinking about the link speeds. Bits and Bytes will bit you in the butt! 🙂
OVH Elite VPS 2023
Cost: $62.70/month USD (No commitment)
Hardware
- OVH VPS vps2020-elite-8-16-320 os-bhs2
- Intel Core (Haswell, no TSX) 2399 MHz (8 cores)
- Network Interface: 1G advertised, 1000 Ethernet via
ip addr
- Location: Beauharnois, Quebec, Canada, in datacenter BHS2.
CPU Test Results
PerformanceTest Result (cpubenchmark.net):
- https://www.passmark.com/baselines/V10/display.php?id=502920511794
- CPU Mark: 6728
- Single Thread: 1365
Geekbench (browser.geekbench.com)
- https://browser.geekbench.com/v5/cpu/19455230
- Single-Core: 556
- Muti-Core: 2985
Network Test Results
- The location of this server is Beauharnois, Quebec, Canada in OVH’s Beauharnois (BHS2) Canada datacenter.
- The network interface is 1G
Speedtest.net
Toronto, Rogers Wireless (Toronto, ON, Canada) [3.18 km](46759)
- ./speedtest-cli –server 46759
- https://www.speedtest.net/result/14204199714.png
Fibrenoire Internet (Montréal, QC, Canada) [504.47 km] (911)
- ./speedtest-cli –server 911
- https://www.speedtest.net/result/14212943059.png
Rogers (Etobicoke, ON, Canada) [11.47 km]
- ./speedtest-cli –server 46144
- https://www.speedtest.net/result/14204222171.png
iperf3 Test
- Used the list at https://github.com/R0GGER/public-iperf3-servers
- New York – iperf3 -c nyc.speedtest.clouvider.net -p 5200-5209 – (Test Server is Listed as having a 10Gbit/s Connection)
- Availabe as IPv6, resolving to IPv4, nyc.speedtest.clouvider.net = 94.154.159.137
Download Test IPv4 – OVH VPS
Upload Test IPv4 – OVH VPS
Download Test IPv6 – OVH VPS
Upload Test IPv6 – OVH VPS
OVH Dedicated Server Advanced LE
Cost: $103.99/month USD (No commitment)
Hardware
- CPU : AMD Ryzen 5 5600X – 6c/12t – 3.7 GHz/4.6 GHz
- Motherboard: ASRockRack X470D4U2-2T
- 2 x RJ45 10G base-T by Intel® X550-AT2
- 1 x RJ45 Dedicated IPMI LAN port by RTL8211
- Memory : 32 GB DDR4 ECC
- Manufacturer: Samsung
- Serial Number: 021D2043
- Part Number: M391A2G43BB2-CWE
- Storage : SSD NVMe
- SAMSUNG MZQLB960HAJR-00007
- Samsung MZQLB960HAJR-00007 960GB PM983 Pci-E 3.0 X4 2.5inch NVMe SSD
- Network
- 2 x RJ45 10G base-T by Intel® X550-AT2 –
- Interface: 1G, 1000 Ethernet via
ip addr
- 1 x RJ45 Dedicated IPMI LAN port by RTL8211
Location and Bandwidth
- Public bandwidth : 1 Gbps
- Private bandwidth : 1 Gbps unmetered
- Location: Beauharnois, Quebec, Canada in datacenter BHS8.
CPU Test Results
PerformanceTest Result (cpubenchmark.net):
- https://www.passmark.com/baselines/V10/display.php?id=502919736013
- CPU Mark: 23,005
- Single Thread: 3459
Geekbench (browser.geekbench.com)
- https://browser.geekbench.com/v5/cpu/19676236
- Single-Core Score: 1734
- Multi-Core Score: 8528
Network Test Results
- The location of this server is Beauharnois, Quebec, Canada in OVH’s Beauharnois (BHS8) – Canada datacenter.
- The network interface is connected at 1000Mbps
- Gigabit Ethernet/GigE = 1Gbps = 1000Mbps
- Mbits = Mbps when using iperf3
Speedtest.net (speedtest-cli)
Rogers Wireless (Toronto, ON, Canada) [3.18 km]
- ./speedtest-cli –server 46759
- https://www.speedtest.net/result/14203747295.png
Fibrenoire Internet (Montréal, QC, Canada) [504.47 km] (911)
- ./speedtest-cli –server 911
- https://www.speedtest.net/result/14212904812.png
Rogers (Etobicoke, ON, Canada) [11.47 km]
- ./speedtest-cli –server 46144
- https://www.speedtest.net/result/14203862278.png
iperf3 Test
Used the list at https://github.com/R0GGER/public-iperf3-servers
New York
- iperf3 -c nyc.speedtest.clouvider.net -p 5200-5209 – (Test Server is Listed as having a 10GB/s Connection)
- Availabe as IPv6, resolving to IPv4, nyc.speedtest.clouvider.net = 94.154.159.137
Download Test IPv4
Download Test IPv6
Upload Test IPv4
Upload Test IPv6
OVH Dedicated Ryzen + Proxmox VM Test Results
I setup Proxmox and created a 4C/8GB virtual instance, and ran PerformanceTest
PerformanceTest Result (cpubenchmark.net)
- https://www.passmark.com/baselines/V10/display.php?id=502983243450
- Single-Core Score: 7,107
- MultiCore Score: 3,158
Final Thoughts
The tests speak for themselves; the $103/month Ryzen outperforms the $62/month VPS substantially.
The strangest but make sense item here is the IPv6 speeds on the VPS. Which are insane compared to the IPv4. The proxmox instance will get IPv6 soon and I can test further. See Testing Notes at the top of this page.
Changelog
- 01-16-2023 – Incorrectly using Google Search caused doubt as to the reporting of Ookla, see notes at the top of the page.
- 01-15-2023 – More research into Ookla speed tests and potential discrepancies in unit conversion.
- 01-15-2023 – Added more details and information about hardware for the OVH dedicated server. Also updated networking images and corrected some mistakes. Included information about iperf3 going over 1Gbps on tests on the OVH VPS.
- 01-13-2023 – Added network testing.