[linux-lvm] snapshots+highmem+'md raid5'

[Heinz J . Mauelshagen]

On Sat, Jun 14, 2003 at 10:07:54PM +0200, Herbert Poetzl wrote:
> On Sat, Jun 14, 2003 at 02:45:11PM +0200, Svetoslav Slavtchev wrote:
> > Quoting Herbert Poetzl <herbert at 13thfloor.at>:
> > 
> > > didn't say that this would be bad, only that
> > > it live-locked my system at high load ...
> > > the tenor on the list was: "highmem is bad"
> > > so I turned it off, and, voila the system was
> > > stable again ...
> > 
> > snapshots + md + highmem is very bad
> > if you don't use snapshots its perfectly OK
> 
> I have to disagree, live-lock under high load is
> not perfectly OK, at least not for me ... 8-)
> 
> > > another, yet unexplained issue, is that on
> > > another system, 4 SCSI discs, form a RAID 5
> > > array (software) which gives about 35MB
> > > throughput, but lvm 1.0.7 ontop of that md
> > > only has about 18MB throughput ...
> > 
> > I've a bit defferent figures from a onboard HPT374 4 ide chanels with
> > 4 IBM deskstars 80Gb, but i still lose 8-15Mb/s compared to plain md :
> > 
> > /dev/hda:
> >  Timing buffer-cache reads:   128 MB in  0.41 seconds =308.43 MB/sec
> >  Timing buffered disk reads:  64 MB in  1.42 seconds = 45.07 MB/sec
> > [root at svetljo root]# hdparm -tT /dev/md13
> > 
> > /dev/md13:
> >  Timing buffer-cache reads:   128 MB in  0.41 seconds =316.05 MB/sec
> >  Timing buffered disk reads:  64 MB in  1.08 seconds = 59.26 MB/sec
> > [root at svetljo root]# hdparm -tT /dev/data
> > data   data2
> > [root at svetljo root]# hdparm -tT /dev/data2/Movies
> > 
> > /dev/data2/Movies:
> >  Timing buffer-cache reads:   128 MB in  0.41 seconds =316.05 MB/sec
> >  Timing buffered disk reads:  64 MB in  1.32 seconds = 48.30 MB/sec
> 
> I posted my results a month ago, but I digged them
> out again, because together with your results, it
> sheds new light on that issue ...
> 
> mine          	raw disc    md(raid5)   lvm on md
> -----------------------------------------------------
> hdparm          14694.40    34129.92    18667.52
> dd 1024k        14988.22    34732.56    18647.98
> dd 32k          15516.06    33945.48    18862.67
> 
> yours          	raw disc    md(raid5)   lvm on md
> -----------------------------------------------------
> hdparm          46151.68    60682.24    49459.20 
> 
> in both cases the lvm lies about 3MB over the raw
> disc throughput, regardless of the md values ...

The reason probably is LVMs read ahead.

> 
> maybe this could lead to some explanation of this
> issue ... unfortunately there was no reply to my
> posting over the last month *sigh*
> 
> best,
> Herbert
> 
> > time dd if=/dev/zero of=/mnt/Movies/BigFile.tm bs=1M count=4000
> > 4000+0 records in
> > 4000+0 records out
> > 0.02user 15.57system 1:37.10elapsed 16%CPU (0avgtext+0avgdata 0maxresident)k
> > 0inputs+0outputs (148major+40minor)pagefaults 0swaps
> > 
> > best,
> > 
> > Svetljo
> > 
> > 
> > _______________________________________________
> > linux-lvm mailing list
> > linux-lvm at sistina.com
> > http://lists.sistina.com/mailman/listinfo/linux-lvm
> > read the LVM HOW-TO at http://tldp.org/HOWTO/LVM-HOWTO/

-- 

Regards,
Heinz    -- The LVM Guy --

*** Software bugs are stupid.
    Nevertheless it needs not so stupid people to solve them ***

=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-

Heinz Mauelshagen                                 Sistina Software Inc.
Senior Consultant/Developer                       Am Sonnenhang 11
                                                  56242 Marienrachdorf
                                                  Germany
Mauelshagen at Sistina.com                           +49 2626 141200
                                                       FAX 924446
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-