3 ms·

A good trick to combine with this is ControlSockets.. you can have multiple SSH/SCP connections over the same actual SSH connections. And starting a new SSH ove

A good trick to combine with this is ControlSockets.. you can have multiple SSH/SCP connections over the same actual SSH connections. And starting a new SSH over the existing connection is much faster than the initial connection particularly if you are on higher latency (e.g. from Australia at 250-400ms) ``` Host xyz HostName 1.2.3.4 ControlMaster auto ControlPath ~/.ssh/sockets/%r@%h-%p ControlPersist 600 ``` Then if you SSH to the same host multiple times it will re-use the connection and it persists for 10 minutes after you disconnect. If you are using a ProxyJump like the above post this can speed up the initial ProxyJump connection.. or you can just use it with normal SSH (which is what I do) and when i want to open multiple tabs to the same machine its significantly faster.

mmalone 7y ago
I've always wondered what this is useful for. So it's purely for performance? Why is it faster to establish the connection? Does it re-use the authentication from the existing ControlMaster too, so you skip the handshake? Seems like it could be dangerous if you're not careful. I guess that's why you can configure it to use `ssh-askpass` for conformation (which, btw, is missing on macOS these days). The other neat-looking directive that I've never tried is `ProxyUseFdpass`, which tells OpenSSH to expect a file descriptor back from your `ProxyCommand` instead of using stdin/stdout. I'm not sure why it exists, but it feels like it could be a performance optimization. Particularly for `scp`. But I've never actually run into a performance problem using `ProxyCommand` so shrug?
- kbenson 7y ago
  I think it's purely for performance, but it does make a big difference. There's a lot of round trips to set up an SSH connection, and depending on the RTT to the server, that can feel like a real delay if there's another host in the connection. If you use a Proxy host for jumping, and it's got a better RTT to most servers you connect to, reusing the existing connection to it might actually make it feel like it connects faster than a direct connection from your workstation would. e.g. $ ping -c 2 foo PING foo (a.b.c.d) 56(84) bytes of data. 64 bytes from foo (a.b.c.d): icmp_seq=1 ttl=55 time=3.56 ms 64 bytes from foo (a.b.c.d): icmp_seq=2 ttl=55 time=3.62 ms --- foo ping statistics --- 2 packets transmitted, 2 received, 0% packet loss, time 3ms rtt min/avg/max/mdev = 3.564/3.594/3.624/0.030 ms $ time ssh root@foo "echo foo" foo real 0m0.650s user 0m0.062s sys 0m0.006s Now if I add the following to .ssh/config: Host foo controlmaster auto controlpath ~/.ssh/ssh-%r@%h:%p You can see how much faster it gets if I already have a connection to that server open: $ time ssh root@foo "echo foo" foo real 0m0.032s user 0m0.003s sys 0m0.004s
  - kqr 7y ago
    This is especially important in some cases where you have local shell wrappers around commands that execute on a remote server with SSH. I do this for my email, for example, where the actual programs for dealing with my email sits on a server, but on each host I have shell scripts with the same name as the original program, except all they do is SSH to the server and run the actual program there. When navigating a GUI that runs these shell wrappers, you get a lot of opening and closing of SSH sessions very quickly.
- Piskvorrr 7y ago
  It's not even establishing the connection: it just goes through the existing ControlMaster. This gets me about ~0.5 sec for a fresh connection, and just-about-instant response for a multiplexed one.
- beagle3 7y ago
  > So it's purely for performance? It's not just for performance - it "reuses authentication" by not needing it. The authentication is an encapsulation, within which multiple channels get used. Usually, it's just the main "shell" channel; but occasionally it's also forwarding (-L / -R / -D), sftp; what ControlMaster does is let you open more "shell" channels when you run the "ssh" command again, instead of establishing a whole new connection. > The other neat-looking directive that I've never tried is `ProxyUseFdpass`, It's so you can use whatever weird communication hardware/protocol you like (serial? non-tcp satellite modem?), and all you have to do to get all of ssh's wonderful features over it, is make provide a file descriptor that works well enough (serial port; pipe you create; etc.)
- pletnes 7y ago
  Also for tab completion of directories to rsync on the remote. Otherwise you’d have to type your password on every <tab>. You can use ssh keys and agents to achieve the same, but it’s much slower, and harder to set up IMO.

mmalone 7y ago

I've always wondered what this is useful for. So it's purely for performance? Why is it faster to establish the connection? Does it re-use the authentication from the existing ControlMaster too, so you skip the handshake? Seems like it could be dangerous if you're not careful. I guess that's why you can configure it to use `ssh-askpass` for conformation (which, btw, is missing on macOS these days). The other neat-looking directive that I've never tried is `ProxyUseFdpass`, which tells OpenSSH to expect a file descriptor back from your `ProxyCommand` instead of using stdin/stdout. I'm not sure why it exists, but it feels like it could be a performance optimization. Particularly for `scp`. But I've never actually run into a performance problem using `ProxyCommand` so shrug?

kbenson 7y ago
I think it's purely for performance, but it does make a big difference. There's a lot of round trips to set up an SSH connection, and depending on the RTT to the server, that can feel like a real delay if there's another host in the connection. If you use a Proxy host for jumping, and it's got a better RTT to most servers you connect to, reusing the existing connection to it might actually make it feel like it connects faster than a direct connection from your workstation would. e.g. $ ping -c 2 foo PING foo (a.b.c.d) 56(84) bytes of data. 64 bytes from foo (a.b.c.d): icmp_seq=1 ttl=55 time=3.56 ms 64 bytes from foo (a.b.c.d): icmp_seq=2 ttl=55 time=3.62 ms --- foo ping statistics --- 2 packets transmitted, 2 received, 0% packet loss, time 3ms rtt min/avg/max/mdev = 3.564/3.594/3.624/0.030 ms $ time ssh root@foo "echo foo" foo real 0m0.650s user 0m0.062s sys 0m0.006s Now if I add the following to .ssh/config: Host foo controlmaster auto controlpath ~/.ssh/ssh-%r@%h:%p You can see how much faster it gets if I already have a connection to that server open: $ time ssh root@foo "echo foo" foo real 0m0.032s user 0m0.003s sys 0m0.004s
- kqr 7y ago
  This is especially important in some cases where you have local shell wrappers around commands that execute on a remote server with SSH. I do this for my email, for example, where the actual programs for dealing with my email sits on a server, but on each host I have shell scripts with the same name as the original program, except all they do is SSH to the server and run the actual program there. When navigating a GUI that runs these shell wrappers, you get a lot of opening and closing of SSH sessions very quickly.
Piskvorrr 7y ago
It's not even establishing the connection: it just goes through the existing ControlMaster. This gets me about ~0.5 sec for a fresh connection, and just-about-instant response for a multiplexed one.
beagle3 7y ago
> So it's purely for performance? It's not just for performance - it "reuses authentication" by not needing it. The authentication is an encapsulation, within which multiple channels get used. Usually, it's just the main "shell" channel; but occasionally it's also forwarding (-L / -R / -D), sftp; what ControlMaster does is let you open more "shell" channels when you run the "ssh" command again, instead of establishing a whole new connection. > The other neat-looking directive that I've never tried is `ProxyUseFdpass`, It's so you can use whatever weird communication hardware/protocol you like (serial? non-tcp satellite modem?), and all you have to do to get all of ssh's wonderful features over it, is make provide a file descriptor that works well enough (serial port; pipe you create; etc.)
pletnes 7y ago
Also for tab completion of directories to rsync on the remote. Otherwise you’d have to type your password on every <tab>. You can use ssh keys and agents to achieve the same, but it’s much slower, and harder to set up IMO.