Performance Benchmark SKUDONET Enterprise Edition v10

Benchmark Scenario

The scenario is created with 2 connected 1GbE networks using the Load Balancer 7108-Mid-size Network Appliance (SNA7108) hardware with the latest SKUDONET Enterprise Edition 10 package optimized for SNA7108.

The networking environment is:

1. 1. Service Network: 172.16.1.0/24
      • Eth4: 172.16.1.1 in SNA Virtual Services
      • Clients: 172.16.1.0/24
      • Virtual Load Balancing: 172.16.1.1:80 (HTTP) and 172.16.1.1:443 (HTTPS)
   2. Backends Network: 172.16.2.0/24
      • Eth5: 172.16.2.1 in SNA
      • Backends: 172.16.2.0/24

The SNA7108 provides a SKUDONET Enterprise Edition 10 version with a CPU Intel® Core™ i5-9500, base Frequency 3.00 GHz with 6 cores and 8GB DDR4 2667 MHz memory (expandable up to 32GB).

The clients and backends host limits haven’t been reached during the benchmarking tests.

The client stress application used was ‘wrk’ which is a HTTP benchmarking tool that generates heavy HTTP and HTTPS requests against the listening service of the load balancer. The backends were configured with a simple ‘nginx’ web service that received the client requests and responded an empty web page (0 bytes of html body and 233 bytes in HTTP headers).

2. Benchmark Cases

2.1. L4xNAT Profile

Farm Configuration in SKUDONET

Command executed in the client

root@client:/# ./wrk -d30 -t10 -c360 http://172.16.1.1/
The execution time was 30 secs long and simulates 10 threads with a concurrency of 360 each, the result is shown below:

Running 30s test @ http://172.16.1.1
10 threads and 360 connections
Thread Stats   Avg      Stdev     Max   +/- Stdev
Latency        1.63ms   0.95ms   51.74ms   93.82%
Req/Sec        47.92k   6.28k   105.01k   85.99%
14325883 requests in 30.10s, 3.33GB read
Requests/sec: 475983.97
Transfer/sec: 113.46MB

CPU usage in SNA7108

While the test was running, the CPU in the SNA7108 was used as follows:

As it’s shown above, all CPUs reached an average of 27,7% and were moving more than 475,000 request per seconds, HTTP requests are managed smoothly.

2.2. HTTP Profile with HTTP Listener

Farm Configuration in SKUDONET

Command executed in the client

root@client:/# ./wrk -d30 -t10 -c360 http://172.16.1.1/
The execution time was 30 secs long and simulates 10 threads with a concurrency of 360
each, the result is shown below:

Running 30s test @ http://172.16.1.1
10 threads and 360 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 13.01ms 27.30ms 512.04ms 88.70%
Req/Sec 10.20k 1.96k 28.03k 82.04%
3043984 requests in 30.08s, 725.60MB read
Requests/sec: 101198.57
Transfer/sec: 24.12MB

CPU usage in SNA7108

While the test is running, the CPU in the SNA7108 is used as follows:

As it’s show, all CPUs of SNA7108 reach the 100% of usage but the HTTP requests are managed smoothly.

2.3. HTTP Profile with HTTPS Offload Listener

Farm Configuration in SKUDONET

Command executed in the client

root@client:/# ./wrk -d30 -t10 -c360 https://172.16.1.1/
The execution time is 30 secs and simulates 10 threads with a concurrency of 360, the output result is shown below:

Running 30s test @ https://172.16.1.1
10 threads and 360 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 12.81ms 57.26ms 1.98s 96.95%
Req/Sec 9.42k 1.82k 31.24k 74.65%
2805269 requests in 30.09s, 789.09MB read
Requests/sec: 93238.41
Transfer/sec: 26.23MB

CPU usage in SNA7108

While the test is running, the CPU in the SNA7108 is used as follows:

As it’s shown above, all CPUs of SNA7108 reach the 100% of usage but the SSL requests are managed smoothly.

3. Results Summary

The following shows the performance benchmark results for all cases over the SNA7108 hardware appliance and web traffic:

* CPU was never saturated (~27%)
**Theoretical values using all CPU resources; 1 RPS is able to simulate 10 concurrent users;

To sum up, in the table above we can see how the SNA7108 load balancer is able to manage more than 475k request per second with a L4xNAT profile using just the 27.7% of the CPU performance. With such performance, the L4xNAT is able to handle more than 17.1 millions of web users per second.

The performance reached with a HTTP farm profile is able to handle more than a million of web users per second and more than 932 thousands of web users per second with high security ciphers in HTTPS layer.