openssh 8.1p1-1 performance regression

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|

openssh 8.1p1-1 performance regression

Frank Eske
I recently updated my Cygwin installation and ran into a strange performance
problem. At least part of the problem was introduced between openssh 8.1p1-1
and 8.0p1-2.

Here's one way to reproduce the problem:
1) Start Cygwin X using "xinit something -- -clipboard"
   File something contains at least:
       xterm &
       exec fvwm

2) Connect to a remote host. The remote host can be running Linux or Windows
   Cygwin with sshd server. The connect command sequence is:
   xhost +machine
   xterm -e ssh -Y -o ServerAliveInterval=10 machine -l userid &
       Where machine is the remote machine name and userid the remote
userid.
   Call this xterm window "window 1."

   From window 1, use "xterm &" to create another remote terminal window.
   Call this second remote xterm window "window 2."

3) Run a simple performance test:
   a) Repeatedly hit the enter key until the window only contains command
      prompts.
   b) Enter "#####" (or anything else, it's just a dummy command)
   c) Hold the enter key and note the speed that the dummy command scrolls
up
      the windows. Window 1 scrolls up rapidly. Window 2 scrolls up slowly.

Notes:
   It also takes longer for window 2 to appear than it used to.

   Try "man bash" or "view some-reasonably-sized-text-file". Try page
up/down
   or arrow up/down. Somewhat surprisingly, there is no detectable
difference
   between the windows. Both have zero latency.

4) Modify your Cygwin installation, back-leveling openssh to 8.0p1-2, and
repeat
   the first performance test. There's no difference between the windows,
and
   the second window start up time improves.

I might not have noticed this problem except that I use a (self-written)
program that uses ncurses for screen update operations. It performed way
worse
than the command roll-up. However, even with the back level openssh, some
latency exists where I don't think any existed before. I haven't found any
Cygwin-related reason for this. I looked at packages xorg-server, xinit, and
xterm, finding no performance changes when back-leveling them. Package
ncurses
was unchanged since my last update.

--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple