QNX RTOS v4 Knowledge Base
QNX RTOS v4 Knowledge Base
| Title |
Phindows Speed-related issues |
| Ref. No. |
QNX.000009523 |
| Category(ies) |
Configuration |
| Issue |
1: Moving windows around within the Phindows session slows down the display unacceptably.
2: Using a Phindows link seems much slower than viewing the same node via a Photon jumpgate.
|
| Solution |
Issue 1: Moving windows around within the Phindows session slows down the display unacceptably.
These symptoms match a situation where Phindows/Photon "thinks" your link is faster than it really is. To check, examine the output of the "set" command from within a pterm/shell window -- what is the BAUD variable set to?
In such a case, you may want to feed Phindows a baud rate with the -b option. You might also try playing with the mouse holdoff parameters (-H option) to vary the amount of data travelling both ways on the link when the mouse is moved.
= = = = = = = = = = = = = = = = = = = = = = = = = = = =
Issue 2: Using a Phindows link seems much slower than viewing the same node via a Photon jumpgate.
Using "jumpgates" (equivalent to phditto -nXXX on a QNX lan) will always be "enormously" faster than using phditto via TCP/IP or serial because phrelay and phditto have been optimized to use native QNX networking as the baseband communications. This includes "hopping" phrelay to the appropriate node to maximize performance.
IPC between processes (ie phrelay and phditto) using QNX messaging achieves rates approaching 98% of the wire. When messages have to be routed through messages to other servers (ie: Socket for TCP/IP), and the extra layer of protocols and holdoffs/acknowledgements has to be taken into account, you would expect overall throughput to be significantly worse (at least 4 times, maybe 10 times slower).
If the Socket server is on a 3rd node, then performance "should" take at least another 2x hit.
What this "should" mean is that if you have a QNX lan with 2 machines running Photon (say nodes 1 & 2), then "phditto -n2" from the Photon session on node 1 "should" notice "significantly" more throughput than "phditto -tx.x.x.2" would experience (same hardware, same wire, same programs but significantly more steps in gettimg probably many more pieces between the applications).
Even though slower than native networking, the QNX implementation of TCP/IP performs better than standard TCP/IP implementations provided on Windows platforms (maybe 2-3x faster... we measured 145KB/sec QNX-Win95, compared to 450KB/sec QNX-QNX on the same machines+network), so using the same machines, but running Windows instead of QNX, and using phindows instead of phditto, you "should" see a further reduction in throughput (but not as great) although the end result would be hard to calibrate since overall throughput would likely be affected greatly by other differences (such as speed of graphics drivers).
x09 |
|
