Re: 1394 Performance

On 4 Feb, Mark Knecht wrote:
>> Perhaps Windows optimizes the gap count. Could that make a
>> notable difference in a one host <-> one disk environment?
>
> Absolutely! (And I'm sure your guess is correct. I know M$ optimizes gap
> count.) It could make a big difference, especially if your transfers are
> moving smaller packets.
[...lots of interesting stuff deleted, thank you...]
>
> Is it not possible to place a Linux box and a Windows box on the same 1394
> bus, let Windows optimize the gap count, then tell Windows to dismount the
> drive, mount it under Linux and do the tests with the gap count the same?
> (Or even use gscanbus or something else to just tell us what the gap count
> was?)

It happens I have winDOS 98SE here to join in a chain. Certainly
not the highest end of Microsoft's product range but worth a try.
The experiment:
- Attach a 120GB Icecube to a 266MHz Linux notebook
- Run hdparm -tT three times and calculate the average:

 Timing buffer-cache reads: 28.6 MB/sec
 Timing buffered disk reads: 8.5 MB/sec

- gscanbus confirms a gap count of 63, of course.
- Connect the windows box, syslog shows among others:
    kernel: ieee1394: Current remote IRM is not 1394a-2000 compliant, resetting...
    last message repeated 5 times
    kernel: ieee1394: Stopping reset loop for IRM sanity
  So Windows tried hard to become IRM and finally won. I guess it
  is also bus manager since it is capable to be according to
  /proc/bus/ieee1394/devices.

- Run hdparm three times again:

 Timing buffer-cache reads: 28.3 MB/sec
 Timing buffered disk reads: 12.0 MB/sec

- But: gscanbus still claims the gap count is 63! Puzzling.
- After Windows is disconnected again, hdparms shows the same bad
  read performance as in the first 3 runs.

Conclusion: The mere hand-holding by a Windows box speeds Linux'
sbp2 up by 40%. I would not believe that if I hadn't measured it
myself. :-)

-- 
Stefan Richter
-=====-=-=-- --=- --=-=
http://arcgraph.de/sr/
-------------------------------------------------------
The SF.Net email is sponsored by EclipseCon 2004
Premiere Conference on Open Tools Development and Integration
See the breadth of Eclipse activity. February 3-5 in Anaheim, CA.
http://www.eclipsecon.org/osdn
_______________________________________________
mailing list Linux1394-user@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux1394-user