I am trying to implement Mark Russinovich's SysInternals PsPing tool in C# to measure latencies using TCP ping.
I am not sure how it makes the ping call (apparently not uses raw Window sockets, because it does not require Administrator privileges to run). I know hping
sends SYN packet over TCP and measures response time.
What would be the implementation technique for accurate measurement of server latency which over TCP does not measure page load times but just the network latency for the package acknowledgement? Is there a library for this?
C:\>psping stackoverflow.com:80
PsPing v2.01 - PsPing - ping, latency, bandwidth measurement utility
Copyright (C) 2012-2014 Mark Russinovich
Sysinternals - www.sysinternals.com
Pinging 198.252.206.16 with 32 bytes of data:
5 iterations (warmup 1) ping test:
Reply from 198.252.206.16: 81.57ms
Reply from 198.252.206.16: 80.81ms
Reply from 198.252.206.16: 80.68ms
Reply from 198.252.206.16: 80.52ms
Reply from 198.252.206.16: 80.71ms
Ping statistics for 198.252.206.16:
Sent = 4, Received = 4, Lost = 0 (0% loss),
Minimum = 80.52ms, Maximum = 80.81ms, Average = 80.68ms
Update: Please don't answer like "why don't you use Ping class, it already does the thing" as I'm asking about ping over TCP, not ICMP.
I have tried several approaches, first thinking I had to use raw sockets or at least use native calls, but a simple TCP connect and close seems to create exactly the same results as the psping utility:
var times = new List<double>();
for (int i = 0; i < 4; i++)
{
var sock = new Socket(AddressFamily.InterNetwork, SocketType.Stream, ProtocolType.Tcp);
sock.Blocking = true;
var stopwatch = new Stopwatch();
// Measure the Connect call only
stopwatch.Start();
sock.Connect(endPoint);
stopwatch.Stop();
double t = stopwatch.Elapsed.TotalMilliseconds;
Console.WriteLine("{0:0.00}ms", t);
times.Add(t);
sock.Close();
Thread.Sleep(1000);
}
Console.WriteLine("{0:0.00} {1:0.00} {2:0.00}", times.Min(), times.Max(), times.Average());
Inspecting the traffic using Wireshark, I can confirm both psping and the snippet above are creating exactly the same sequence of packets.
-> [SYN]
<- [SYN,ACK]
-> [ACK]
-> [FIN,ACK]
<- [FIN,ACK]
-> [ACK]
Output from psping with no warm-up and using TCP ping:
C:\>psping -w 0 stackoverflow.com:80
PsPing v2.01 - PsPing - ping, latency, bandwidth measurement utility
Copyright (C) 2012-2014 Mark Russinovich
Sysinternals - www.sysinternals.com
TCP connect to 198.252.206.16:80:
4 iterations (warmup 0) connecting test:
Connecting to 198.252.206.16:80: 92.30ms
Connecting to 198.252.206.16:80: 83.16ms
Connecting to 198.252.206.16:80: 83.29ms
Connecting to 198.252.206.16:80: 82.98ms
TCP connect statistics for 198.252.206.16:80:
Sent = 4, Received = 4, Lost = 0 (0% loss),
Minimum = 82.98ms, Maximum = 92.30ms, Average = 85.43ms
Output from the program above:
C:\>TcpPing.exe stackoverflow.com 80
88.60ms
83.65ms
84.05ms
84.05ms
83.65 88.60 85.09
As for measurements, I must say, sometimes there are quite a few different results at different runs, but overall they seemed pretty close: kind of proves measuring the Connect()
call is good enough. I'm thinking, taking a median of a few hundred results might prove it with a bit more confidence.