βA picture stolen from the west to attract attentionβ
In our previous articles, we talked about how to work with on Windows Server Core 2019 on our new UltraLight plan for 99 rubles per month. We offer another way to use this tariff. This time we will talk about what is better to choose if you need a VPN for lazy people or a static IP address, which, by the way, is more convenient to use instead of Hamachi and everything else, if you really want to play heroes or Warcraft 3 on a local network. It's not about tuning, let's talk about performance.
Testing technique
RRAS and SoftEther were chosen based on ease of installation, L2TP protocol support, and GUI control.
SoftEther and RRAS used an L2TP connection with a shared key through standard Windows tools. As established, so it was tested.
The operating system for SoftEther is Ubuntu 18.04 LTS, for RRAS Windows Server Core 2019. Before the tests, all operating systems received the latest updates as of 21.11.2019/XNUMX/XNUMX.
1 GB of RAM was installed on the second generation Hyper-V virtual machine, as well as limits on the processor. The order in which the test groups are run is as follows:
For all 8 cores:
- Without restrictions
- 50% limit
- 25% limit
- 5% limit
- 1% limit
For 4 cores:
- Without restrictions
- 50% limit
- 25% limit
- 5% limit
- 1% limit
For one core:
- Without restrictions
- 50% limit
- 25% limit
- 5% limit
- 1% limit
All VPN servers used out of the box settings and had NAT enabled. All virtual machines are on the same host and on the same virtual switch.
To evaluate network performance, testing was performed between the server and the client without a VPN connection.
The test was carried out using TamoSoft Throughput Test in TCP only mode, for tables and graphs, the values ββ\u5b\u30bof "ave" were taken. Data was collected for XNUMX minutes XNUMX seconds for each test.
For a better understanding of the limits of both implementations, we first test the throughput of the virtual switch.
This is what the test results looked like. Further, all results will be wrapped in tables.
As you can see, the virtual switch is not a bottleneck in testing and almost gives out a theoretical limit of 10 gigabits.
How "Physically" the test network looked like
Results:
For one core:
In a single-core discipline, both servers are on a par.
For 4 cores:
For 8 cores:
Here we clearly see which solution scales best with the number of cores. By reducing the performance of each of the cores, RRAS compensated for the loss in their number, which SoftEther did not.
System RAM consumption
The amount of RAM consumed by SoftEther increased from the number of cores, from 122 to 177 MB, but still less than that of RRAS.
The RRAS service itself weighs about 200 megabytes in memory, minus the total system consumption.
Throughput under different conditions
The total throughput without any restrictions on the processor.
If you have not yet chosen the solution that suits you, perhaps this table will help you make your choice. The total throughput in the CPU deficit mode is given.
Please note that on four and one core SoftEther's performance is higher than on eight. Such low performance is not found anywhere else, but the testing itself shows how well the algorithm scales with the number of cores.
Conclusion:
Connecting to SoftEther with a processor limit did not work the first time, I had to first increase the limit, connect and only then lower the limit, this imposes a restriction on its installation in very thin environments. RRAS always logged in instantly.
If you have a machine with a lot of cores, RRAS is the way to go. And for SoftEther'a, you can leave 4 cores. The author, if he had used it, would have left only one core for him.
What and where to place - decide for yourself. If you have 99 rubles for with Windows Server on board, RRAS will still be the best choice.
Source: habr.com