hdparm -tT reports "suspicious results" ???

hdparm -tT reports "suspicious results" ???

[Chuck Ebbert]

  Real-world benchmarks show I am getting about 30-32 MB/sec from my
IDE disks on 2.4.21-pre6: raid1 rebuild reports 32MB/sec and copying
a 1GB file to /dev/null on a newly-mounted fs takes 33 sec.  I have
moved the disk to another computer that is almost 100% identical
except for CPU (PPro 200 vs. PII ODP/333) and now hdparm says 24MB/sec
but I am still getting 30-32 real world on the PPro 200.  The below
results are from the PII Overdrive (I have not tested real-world
on 2.5 but notice how much slower hdparm reports its speed.)  What
does "suspicious" mean here?



# uname -a
Linux d2 2.4.21-pre6 #3 Fri Apr 4 18:22:11 EST 2003 i686 unknown
# hdparm -tT /dev/hde

/dev/hde:
 Timing buffer-cache reads:   128 MB in  2.50 seconds = 51.20 MB/sec
 Timing buffered disk reads:  64 MB in  2.15 seconds = 29.77 MB/sec
Hmm.. suspicious results: probably not enough free memory for a proper test.
# 
Script done on Fri Apr  4 20:49:01 2003
Script started on Fri Apr  4 20:52:03 2003
# uname -a
Linux d2 2.4.20 #1 Fri Apr 4 17:33:07 EST 2003 i686 unknown
# hdparm -tT /dev/hde

/dev/hde:
 Timing buffer-cache reads:   128 MB in  2.62 seconds = 48.85 MB/sec
 Timing buffered disk reads:  64 MB in  2.35 seconds = 27.23 MB/sec
Hmm.. suspicious results: probably not enough free memory for a proper test.
# 
Script done on Fri Apr  4 20:52:46 2003
Script started on Fri Apr  4 20:56:54 2003
# uname -a
Linux d2 2.5.66 #11 Sat Mar 29 21:14:17 EST 2003 i686 unknown
# hdparm -tT /dev/hde

/dev/hde:
 Timing buffer-cache reads:   128 MB in  2.66 seconds = 48.14 MB/sec
 Timing buffered disk reads:  64 MB in  3.90 seconds = 16.41 MB/sec
# cat /proc/meminfo
MemTotal:       191956 kB
MemFree:        170728 kB
Buffers:          4244 kB
Cached:           9548 kB
SwapCached:          0 kB
Active:           7908 kB
Inactive:         7764 kB
HighTotal:           0 kB
HighFree:            0 kB
LowTotal:       191956 kB
LowFree:        170728 kB
SwapTotal:      262040 kB
SwapFree:       262040 kB
Dirty:               8 kB
Writeback:           0 kB
Mapped:           3668 kB
Slab:             3932 kB
Committed_AS:     2616 kB
PageTables:        292 kB
ReverseMaps:      2832


Configuration (now):
 Dell Optiplex GXPro (MPS 1.4)
      1 PPro 200, Uniprocessor IO-APIC kernel
      Promise Ultra 100TX2
              2 x Maxtor 4K060H3 (60GB 5400RPM)
              (1 drive per channel)

--
 Chuck
 I am not a number!

Re: hdparm -tT reports "suspicious results" ???

[Chuck Ebbert]

Mark Hahn wrote:


>does "PII ODP" mean a PII in an adapter that fits into a P6 socket?


It is a factory Intel PII/333 with an HSF that fits in Socket 8.
With its 512K full-speed cache it makes a PPro into a half-decent machine.
Too bad they're so rare...


>> /dev/hde:
>>  Timing buffer-cache reads:   128 MB in  2.62 seconds = 48.85 MB/sec
>
> hmm, slower dram and/or FSB.


No, same exact machine in _every_ way, just a different kernel.  It is
slow ECC EDO memory, though.


>> /dev/hde:
>>  Timing buffer-cache reads:   128 MB in  2.66 seconds = 48.14 MB/sec
>>  Timing buffered disk reads:  64 MB in  3.90 seconds = 16.41 MB/sec.>
>
> dram is still slow, but disk appears slower yet, so not obviously
> contaminating the measurement.

..and that's 2.5 which measures much lower but doesn't seem to be that

slow (still not tested, I'm testing RAID1 patches on 2.4 now.)


> 53073H4, ecs k7s6a-pro, athlon/"1800", DDR:
> /dev/hda:
>  Timing buffer-cache reads:   128 MB in  0.48 seconds =266.67 MB/sec
>  Timing buffered disk reads:  64 MB in  2.04 seconds = 31.37 MB/sec


3 x 33073H3, RAID1, Abit KT7A-RAID, Tbird 1333, PC133 SDRAM:


Personalities : [raid0] [raid1] 
read_ahead 24 sectors
md2 : active raid1 hdg9[0] hde8[2] hda8[1]
      2409600 blocks [3/3] [UUU]

2.4.21-pre6:
/dev/md2:
 Timing buffer-cache reads:   128 MB in  0.58 seconds =220.69 MB/sec
 Timing buffered disk reads:  64 MB in  2.71 seconds = 23.62 MB/sec

2.4.20aa1ce1(unreleased):
/dev/md2:
 Timing buffer-cache reads:   128 MB in  0.59 seconds =216.95 MB/sec
 Timing buffered disk reads:  64 MB in  1.67 seconds = 38.32 MB/sec


I have RAID1 patches that give 55% sequential IO improvements on K7 but
they're not showing the same gains on the PPro (<20% so far.)  Kernel,
chipset, CPU and disk controller are all different, though I expected
to get proportional gains on any hardware.

AFAICT hdparm says "suspicious results" when the disk speed is greater
than one-half the buffer speed.  I just got 51.20 and 26.34 on the PPro
with my patches and the message came out again.

--
 Chuck
 I am not a number!