>>>>> "Izumi" == Izumi Tsutsui <tsutsui@ceres.dti.ne.jp> writes:

    Izumi> Umm, I should have tried benchmark on my RaQ2 first.

    Izumi> It seems setting the GT64x11 timeout and retry register
    Izumi> improves disk I/O performance on my RaQ2 at least with
    Izumi> bonnie:

I did bonnie and ttcp tests on my qube2 with your COBALT_TEST patch
and w/o and w/ your additional gt.c patch (GT64x11 timeout and retry
register).

Result w/o your additional gt.c patch:

  [Sat May 20 12:44:13 cochise ~]$ bonnie 
  File './Bonnie.416', size: 104857600
  Writing with putc()...done
  Rewriting...done
  Writing intelligently...done
  Reading with getc()...done
  Reading intelligently...done
  Seeker 2...Seeker 1...Seeker 3...start 'em...done...done...done...
                -------Sequential Output-------- ---Sequential Input-- --Random--
                -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
  Machine    MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU  /sec %CPU
            100  3695 93.1  4505 70.0  3645 80.2  3004 63.6 14500 97.1 456.9 93.9
  
  [Sat May 20 12:37:18 cochise ~]$ ttcp -rs
  ttcp-r: buflen=8192, nbuf=2048, align=16384/0, port=5001  tcp
  ttcp-r: socket
  ttcp-r: accept from 172.27.64.183
  ttcp-r: 16777216 bytes in 7.30 real seconds = 2243.40 KB/sec +++
  ttcp-r: 2049 I/O calls, msec/call = 3.65, calls/sec = 280.56
  ttcp-r: 0.0user 6.8sys 0:07real 94% 0i+0d 0maxrss 0+2pf 0+472csw
  
  [Sat May 20 12:41:18 cochise ~]$ ttcp -ts qie
  ttcp-t: buflen=8192, nbuf=2048, align=16384/0, port=5001  tcp  -> qie
  ttcp-t: socket
  ttcp-t: connect
  ttcp-t: 16777216 bytes in 8.54 real seconds = 1917.49 KB/sec +++
  ttcp-t: 2048 I/O calls, msec/call = 4.27, calls/sec = 239.69
  ttcp-t: 0.0user 8.1sys 0:08real 95% 0i+0d 0maxrss 0+4098pf 0+541csw

Result w/ your additional gt.c patch:

  [Sat May 20 13:07:50 cochise ~]$ bonnie 
  File './Bonnie.322', size: 104857600
  Writing with putc()...done
  Rewriting...done
  Writing intelligently...done
  Reading with getc()...done
  Reading intelligently...done
  Seeker 1...Seeker 3...Seeker 2...start 'em...done...done...done...
                -------Sequential Output-------- ---Sequential Input-- --Random--
                -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
  Machine    MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU  /sec %CPU
            100  3753 94.4  7220 88.3  5631 92.8  4308 96.1 13014 96.6 691.0 95.9
  
  [Sat May 20 13:18:51 cochise ~]$ ttcp -rs
  ttcp-r: buflen=8192, nbuf=2048, align=16384/0, port=5001  tcp
  ttcp-r: socket
  ttcp-r: accept from 172.27.64.183
  ttcp-r: 16777216 bytes in 7.86 real seconds = 2083.54 KB/sec +++
  ttcp-r: 2049 I/O calls, msec/call = 3.93, calls/sec = 260.57
  ttcp-r: 0.0user 7.1sys 0:07real 91% 0i+0d 0maxrss 0+2pf 0+514csw
  
  [Sat May 20 13:12:07 cochise ~]$ ttcp -ts qie
  ttcp-t: buflen=8192, nbuf=2048, align=16384/0, port=5001  tcp  -> qie
  ttcp-t: socket
  ttcp-t: connect
  ttcp-t: 16777216 bytes in 8.94 real seconds = 1833.07 KB/sec +++
  ttcp-t: 2048 I/O calls, msec/call = 4.47, calls/sec = 229.13
  ttcp-t: 0.0user 8.5sys 0:08real 96% 0i+0d 0maxrss 0+4098pf 0+592csw

So, on my qube2

  NetBSD 3.99.20 (COCHISE) #3: Sat May 20 12:54:04 MEST 2006
          kilbi@qie:/usr/src/sys/arch/cobalt/compile/COCHISE
  Cobalt Qube 2
  total memory = 256 MB
  avail memory = 246 MB
  [...]
  wd0 at atabus0 drive 0: <SAMSUNG SV1604N>
  wd0: drive supports 16-sector PIO transfers, LBA48 addressing
  wd0: 149 GB, 310101 cyl, 16 head, 63 sec, 512 bytes/sect x 312581808 sectors
  wd0: 32-bit data port
  wd0: drive supports PIO mode 4, DMA mode 2, Ultra-DMA mode 5 (Ultra/100)
  wd0(viaide0:0:0): using PIO mode 4, Ultra-DMA mode 2 (Ultra/33) (using DMA)

this additional gt.c patch seems to improve ata/ide disk performance
(partially), but seems to decrease network (tlp0 in my case)
performance.

Markus.