Note also: I have no real explanation for why Case 4 was faster than
Case 5, particularly if that wasn't the case for Case 1 and 2.  However,
one important clue is the fact that VGL_USEXSHM=0 was required for Case
5.  If in fact you were using the VGL Transport with JPEG encoding, then
VGL_USEXSHM could not have had any effect unless you set it on the
client machine (VGL_USEXSHM only affects the X11 Transport or
vglclient.)  I'm not 100% convinced that you were using the VGL
Transport in Case 5.  But regardless, like I said below, running a
window manager with Cases 1, 2, 4, or 5 is not a good idea.

On 3/18/19 10:54 PM, DRC wrote:
> Ah, OK.  It makes much more sense now.  Some notes:
> 
> Case 1, 4:
> 
> This is using the X11 Transport over SSH, which is never a good idea. 
> Even with gigabit, the most you could ever hope to stream with
> uncompressed XPutImage() requests is in the neighborhood of 30
> Megapixels/second (bearing in mind that each pixel is a full 32 bits in
> this case.)  Also, the overhead of SSH is likely going to reduce the
> throughput somewhat because of the immense size of the data stream (and
> thus the CPU cycles required to encrypt it.)  Perceptually lossless JPEG
> (Q95, 4:4:4) over 100 Megabit can easily do > 50 Megapixels/second, by
> comparison.  In short, this configuration is not something that should
> ever be used in production.
> 
> Case 2, 5:
> 
> Assuming you're already running vglclient on the client machine, this is
> the equivalent of using vglconnect (Note that '-c jpeg' is superfluous,
> because VGL will automatically select the VGL Transport w/ JPEG if it
> detects this configuration.)  This configuration should work fine if
> you're running a regular application over the gigabit CAN, but running a
> window manager is a different story.  Window managers do a lot of very
> image-intensive operations and require a lot of back & forth with the X
> server, so in general, it's not a good idea to run them without an X
> proxy.  It could simply be that i3 is less well-tuned for a remote X
> connection-- window managers in general aren't designed to be run over
> remote X.  Even on a gigabit CAN, there is going to be some network
> latency and overhead, and that can add up if the application (the WM, in
> this case) is issuing a lot of fine-grained X requests.
> 
> Cases 1, 2:
> 
> Not sure what you mean by "Cases 1 and 2 with i3 requires XFree86 to be
> disabled."  XFree86 hasn't been used by any Linux distribution in quite
> some time.
> 
> Case 3, 6:
> 
> This is an X proxy, so the actual connection to the X server is entirely
> within the server machine.  Thus, window managers are appropriate in
> this configuration.
> 
> Cases 1, 2, 3:
> 
> VGL_FORCEALPHA was required because of the amdgpu drivers, as previously
> noted.
> 
> Cases 4, 5:
> 
> Still very odd that VGL_USEXSHM=0 was required here, but the fact that
> Cases 1 and 2 didn't require that points to an issue with the server
> machine.
> 
> 
> On 3/18/19 2:47 PM, Jason wrote:
>>
>> Good Afternoon,
>>
>>
>> My mistake, i3 is a window manager (I get them mixed up from time to
>> time :) ); LightDM is the display manager. Here is (to my knowledge)
>> the layout of the environments:
>>
>>
>> VGL_Env.png
>>
>>
>>  *
>>
>>     Cases 1, 2, 4, and 5 do not work well when used with an i3 desktop
>>     (but well for MATE, KDE, etc.).
>>
>>  *
>>
>>     Cases 1 and 2 with i3 requires XFree86 to be disabled
>>
>>  *
>>
>>     Cases 4 and 5 with i3 requires the VGL_USEXSHM=0 to be used
>>
>>  *
>>
>>     Cases 1, 2, and 3 requires the VGL_FORCEALPHA=1 to be used
>>
>>  *
>>
>>     Case 4 seemed to work better than Case 5 in performance
>>     (increasing the compression core count helped, but was still slower)
>>
>>  *
>>
>>     Case 6 works when running vglrun in a terminal, but not when
>>     launching the desktop environment with "vglrun +wm" (audio is lost)
>>
>> (Understandably, cases 3 and 6 are not supported like TurboVNC is, but
>> I included them for completeness sake)
>>
>>
>> In no case is the scope going wider than CAN (while there is potential
>> for WAN, we are focusing currently on the LAN-CAN performance).
>>
>> In all, I am slowly, but surely understanding how Xorg, GLX and SSH
>> all tie together. That being said, I appreciate your time helping us
>> out with all of this.
>>
>>
>> Thanks,
>>
>>    Jason
>>
>> -- 
>> You received this message because you are subscribed to the Google
>> Groups "VirtualGL User Discussion/Support" group.
>> To unsubscribe from this group and stop receiving emails from it, send
>> an email to [email protected]
>> <mailto:[email protected]>.
>> To view this discussion on the web visit
>> https://groups.google.com/d/msgid/virtualgl-users/3174c035-7240-4a74-9b15-860f770f39f0%40googlegroups.com
>> <https://groups.google.com/d/msgid/virtualgl-users/3174c035-7240-4a74-9b15-860f770f39f0%40googlegroups.com?utm_medium=email&utm_source=footer>.
>> For more options, visit https://groups.google.com/d/optout.
> 

-- 
You received this message because you are subscribed to the Google Groups 
"VirtualGL User Discussion/Support" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To view this discussion on the web visit 
https://groups.google.com/d/msgid/virtualgl-users/5ce4238d-a511-2780-1f1a-109e4d40c974%40virtualgl.org.
For more options, visit https://groups.google.com/d/optout.

Reply via email to