SW-RAID0 Performance problems

SW-RAID0 Performance problems

[Andreas Peter]

Hi,
I've successfully set up SW-RAID0 with Kernel 2.4.3 and Raidtools 0.9.
I did this to increase the performance of my HD, but nothig happens.
The hdparm results:
hdparm -t /dev/md0 : 20.25 MB/sec
hdparm -t /dev/hda : 20.51 MB/sec
hdaprm -t /dev/hdc : 20.71 MB/sec

I thougt the performnace of RAID0 should near 40MB/sec.
I played with different chunk-sizes, but the result was everytime the same.
The drives are both Maxtor DiamondMax VL40, 30GB, DMA on. 
No other drive is attached on the bus.

Here are also some bonnie++ results:


-- RAID-0 --
-- chunk-size=16 --

Version  1.01        ------Sequential Output------ --Sequential Input- --Random-
                             -Per Chr- --Block-- -Rewrite-     -Per Chr- --Block-- --Seeks--
Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec %CP
debian           1G     7416    99  14277   20   7498     10   6942    90 27007    20 113.0      1
                       ------Sequential Create------ --------Random Create--------
                      -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
              files   /sec %CP  /sec %CP    /sec %CP   /sec %CP   /sec %CP    /sec %CP
                 16   267    99  +++++ +++ 10968 100   269  99    +++++ +++  1388  99

-- chunk-size=32 --

Version  1.01       ------Sequential Output------ --Sequential Input- --Random-
                    -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec %CP
debian           1G    7396     99 14075    20  7469    10    6945    90   26960  20   133.7   1
                    ------Sequential Create------ --------Random Create--------
                    -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
              files  /sec %CP  /sec %CP   /sec %CP   /sec %CP   /sec %CP    /sec %CP
                 16   265 100 +++++ +++ 10695  99    267   99   +++++ +++  1447 100

-- Single HD /dev/hdc1 --

Version  1.01       ------Sequential Output------ --Sequential Input- --Random-
                           -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP  /sec %CP
debian           1G   7173   96    11055  13     5038   6     5999     78    29146  21 90.7       1
                    ------Sequential Create------ --------Random Create--------
                    -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
              files  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP
                 16   272 100 +++++ +++ 10482  99   274  99 +++++ +++  1437 100

Are there known performanace problems with 2.4.3, or is it necessary to 
apply patches to the kernel? 
Or did I something wrong??
Thank you for every hint!

Andreas
-- 
Andreas Peter *** ujq7@rz.uni-karlsruhe.de

Re: SW-RAID0 Performance problems

[Jakob Østergaard]

On Fri, Apr 13, 2001 at 01:47:30PM +0200, Andreas Peter wrote:
> Hi,
> I've successfully set up SW-RAID0 with Kernel 2.4.3 and Raidtools 0.9.
> I did this to increase the performance of my HD, but nothig happens.
> The hdparm results:
> hdparm -t /dev/md0 : 20.25 MB/sec
> hdparm -t /dev/hda : 20.51 MB/sec
> hdaprm -t /dev/hdc : 20.71 MB/sec
> 
> I thougt the performnace of RAID0 should near 40MB/sec.
> I played with different chunk-sizes, but the result was everytime the same.
> The drives are both Maxtor DiamondMax VL40, 30GB, DMA on. 
> No other drive is attached on the bus.
> 
> Here are also some bonnie++ results:
...

I can't say much about this... It looks like your setup is perfectly allright,
and the performance *should* go up.   Instead it looks like you get a small
performance drop from using the RAID.   Most odd.

Do you have more controllers in the machine ? If so could you try to move eg.
hdc to the second controller ?  The only thing I can imagine being the cause of
the poor performance is, if your controller somehow doesn't handle both
channels very well simultaneously.   It's far fetched, but it's the only
suggestion I can think of.

Maybe Andre has comments ?

I usually get a good speedup from using RAID-0 on 2.4.3 with IDE.  Both with
two disks and with six.   This is with Intel PIIX4 and Promise 20262
controllers.

-- 
................................................................
:   jakob@unthought.net   : And I see the elder races,         :
:.........................: putrid forms of man                :
:   Jakob Østergaard      : See him rise and claim the earth,  :
:        OZ9ABN           : his downfall is at hand.           :
:.........................:............{Konkhra}...............:

Re: SW-RAID0 Performance problems

[Andreas Peter]

Am Freitag, 13. April 2001 18:01 schrieb Jakob Østergaard:

> I can't say much about this... It looks like your setup is perfectly
> allright, and the performance *should* go up.   Instead it looks like you
> get a small performance drop from using the RAID.   Most odd.
>
> Do you have more controllers in the machine ? If so could you try to move
> eg. hdc to the second controller ?  The only thing I can imagine being the
> cause of the poor performance is, if your controller somehow doesn't handle
> both channels very well simultaneously.   It's far fetched, but it's the
> only suggestion I can think of.

The Board ist a Gigabyte 6BXDS (BX-Chipset) and 2 Celeron 533
I think the hardware is ok.

> I usually get a good speedup from using RAID-0 on 2.4.3 with IDE.  Both
> with two disks and with six.   This is with Intel PIIX4 and Promise 20262
> controllers.

Andreas
-- 
Andreas Peter *** ujq7@rz.uni-karlsruhe.de

Re: SW-RAID0 Performance problems

[Andreas Peter]

Mark Hahn schrieb:
> > hdparm -t /dev/md0 : 20.25 MB/sec
> > hdparm -t /dev/hda : 20.51 MB/sec
> > hdaprm -t /dev/hdc : 20.71 MB/sec
>
> md0 is composed of partitions located where on hda and hdc?
> also, what's your CPU?

This is my raidtab file:

raiddev                 /dev/md0
 
raid-level              0    # it's not obvious but this *must* be
                             # right after raiddev
 
persistent-superblock   1    # set this to 1 if you want autostart,
                             # BUT SETTING TO 1 WILL DESTROY PREVIOUS
                             # CONTENTS if this is a RAID0 array created
                             # by older raidtools (0.40-0.51) or mdtools!
 
chunk-size              32
 
nr-raid-disks           2
nr-spare-disks          0
 
device                  /dev/hda3
raid-disk               0
 
device                  /dev/hdc3
raid-disk               1

The partition table:

Disk /dev/hda: 16 heads, 63 sectors, 59556 cylinders
Units = cylinders of 1008 * 512 bytes
 
   Device Boot    Start       End    Blocks   Id  System
/dev/hda1   *         1        21     10552+  83  Linux
/dev/hda2            22       542    262584   82  Linux swap
/dev/hda3           543     59556  29743056   fd  Linux raid autodetect

My board is a Gigabyte 6BXDS BX-Chipset 2 Celerons@533 MHz

Andreas
-- 
Andreas Peter *** ujq7@rz.uni-karlsruhe.de

Re: SW-RAID0 Performance problems

[David Rees]

On Fri, Apr 13, 2001 at 05:36:55PM +0200, Andreas Peter wrote:
> Mark Hahn schrieb:
> > > hdparm -t /dev/md0 : 20.25 MB/sec
> > > hdparm -t /dev/hda : 20.51 MB/sec
> > > hdaprm -t /dev/hdc : 20.71 MB/sec
> >
> > md0 is composed of partitions located where on hda and hdc?
> > also, what's your CPU?
> 
> This is my raidtab file:

<snip>

Cconfig and setup looks OK.

What happens if your run hdparm -t /dev/hda and /dev/hdc at the same time?

-Dave

Re: SW-RAID0 Performance problems

[Andreas Peter]

Am Freitag, 13. April 2001 18:07 schrieb David Rees:

> Cconfig and setup looks OK.
>
> What happens if your run hdparm -t /dev/hda and /dev/hdc at the same time?

Good idea!
The performance is only ~11MB/sec per disk
There is a bottleneck somewhere...

Andreas
-- 
Andreas Peter *** ujq7@rz.uni-karlsruhe.de

Re: SW-RAID0 Performance problems

[David Rees]

On Fri, Apr 13, 2001 at 06:28:20PM +0200, Andreas Peter wrote:
> Am Freitag, 13. April 2001 18:07 schrieb David Rees:
> 
> > Cconfig and setup looks OK.
> >
> > What happens if your run hdparm -t /dev/hda and /dev/hdc at the same time?
> 
> Good idea!
> The performance is only ~11MB/sec per disk
> There is a bottleneck somewhere...

OK, so it's not the RAID setup.  There's two things that can cause this.  
One is that DMA is turned off  (what does hdparm /dev/hda and hdparm 
/dev/hdc show?), the second was that the drives are on the same channel 
(which obviously isn't the case here).  Can you verify that the drives are 
in DMA mode?

-Dave

Re: SW-RAID0 Performance problems

[Andreas Peter]

Am Samstag, 14. April 2001 09:04 schrieb David Rees:

> OK, so it's not the RAID setup.  There's two things that can cause this.
> One is that DMA is turned off  (what does hdparm /dev/hda and hdparm
> /dev/hdc show?), the second was that the drives are on the same channel
> (which obviously isn't the case here).  Can you verify that the drives are
> in DMA mode?

hdparm /dev/hda 

/dev/hda:
 multcount    = 16 (on)
 I/O support  =  0 (default 16-bit)
 unmaskirq    =  0 (off)
 using_dma    =  1 (on)
 keepsettings =  0 (off)
 nowerr       =  0 (off)
 readonly     =  0 (off)
 readahead    =  8 (on)
 geometry     = 59556/16/63, sectors = 60032448, start = 0

the same on /dev/hdc

I played with different hdparm-settings, but it's not possible to speed up 
the HDs

Andreas
-- 
Andreas Peter *** ujq7@rz.uni-karlsruhe.de

Re: SW-RAID0 Performance problems

[Kurt Roeckx]

On Sat, Apr 14, 2001 at 11:38:06AM +0200, Andreas Peter wrote:
> Am Samstag, 14. April 2001 09:04 schrieb David Rees:
> 
> > OK, so it's not the RAID setup.  There's two things that can cause this.
> > One is that DMA is turned off  (what does hdparm /dev/hda and hdparm
> > /dev/hdc show?), the second was that the drives are on the same channel
> > (which obviously isn't the case here).  Can you verify that the drives are
> > in DMA mode?
> 
> hdparm /dev/hda 
> 
> /dev/hda:
>  multcount    = 16 (on)
>  I/O support  =  0 (default 16-bit)
>  unmaskirq    =  0 (off)
>  using_dma    =  1 (on)

Does turning unmaskirq on help?


Kurt

Re: SW-RAID0 Performance problems

[Andreas Peter]

Am Samstag, 14. April 2001 14:28 schrieb Kurt Roeckx:
> Does turning unmaskirq on help?

Already tried this, but it doesn't help
The actual settings (same on /dev/hdc):

bash-2.04# hdparm /dev/hda
 
/dev/hda:
 multcount    = 16 (on)
 I/O support  =  1 (32-bit)
 unmaskirq    =  1 (on)
 using_dma    =  1 (on)
 keepsettings =  0 (off)
 nowerr       =  0 (off)
 readonly     =  0 (off)
 readahead    =  8 (on)
 geometry     = 59556/16/63, sectors = 60032448, start = 0

bash-2.04# hdparm -tT /dev/md0
 
/dev/md0:
 Timing buffer-cache reads:   128 MB in  1.30 seconds = 98.46 MB/sec
 Timing buffered disk reads:  64 MB in  3.14 seconds = 20.38 MB/sec

bash-2.04# hdparm -tT /dev/hda3
 
/dev/hda3:
 Timing buffer-cache reads:   128 MB in  1.31 seconds = 97.71 MB/sec
 Timing buffered disk reads:  64 MB in  2.26 seconds = 28.32 MB/sec

Andreas
-- 
Andreas Peter *** ujq7@rz.uni-karlsruhe.de

Re: SW-RAID0 Performance problems

[Tim Moore]

David Rees wrote:
> What happens if your run hdparm -t /dev/hda and /dev/hdc at the same time?

Try 'hdparm -tT'  with simultaneous /dev/hda3 and /dev/hdc3.  This gives you
a baseline on the actual partitions involved.

rgds,
tim.
--

Re: SW-RAID0 Performance problems

[Andreas Peter]

Am Freitag, 13. April 2001 20:11 schrieb Tim Moore:

> Try 'hdparm -tT'  with simultaneous /dev/hda3 and /dev/hdc3.  This gives
> you a baseline on the actual partitions involved.

hdparm -tT simultanous on /dev/hda3 and /dev/hdc3:

/dev/hda3:
 Timing buffer-cache reads:   128 MB in  2.29 seconds = 55.90 MB/sec
 Timing buffered disk reads:  64 MB in  4.67 seconds = 13.70 MB/sec

/dev/hdc3:
 Timing buffer-cache reads:   128 MB in  2.28 seconds = 56.14 MB/sec
 Timing buffered disk reads:  64 MB in  4.61 seconds = 13.88 MB/sec

Now on single HD  -  /dev/hda3 :

/dev/hda3:
 Timing buffer-cache reads:   128 MB in  1.30 seconds = 98.46 MB/sec
 Timing buffered disk reads:  64 MB in  2.26 seconds = 28.32 MB/sec

It looks like reading on /dev/hda3 locks /dev/hdc3 ...
Is it necessary to apply  the ide-patches to the kernel ?

Andreas
-- 
Andreas Peter *** ujq7@rz.uni-karlsruhe.de