Re: Performance difference between two raid0 arrays on same drives?

Re: Performance difference between two raid0 arrays on same drives?

[Michel Bellais]

You're right, I thought about it too, but the fastest array is built with 
partitions closer to the centre of the disk, so it should be the slowest 
indeed.
The disks are big (180 Gb), the partitions represent less than 10% of it and 
follow each others.  It cannot explain 30% difference in performance.

I have created a third array on the disk, which is a copy of the slowest 
array. It has the same content. This last array shows much better performance 
than the original. And it is even closer to the centre...
So i really don't understand.
I intended to use raid0 to boost my pc at home, so it is not a crucial 
problem, i will use the late array i created.

Thank you for your answer!

Michel Bellais

On Saturday 12 July 2003 06:36 am, Gregory Leblanc wrote:
> On Fri, 2003-07-11 at 09:13, Michel wrote:
> > Hi!
> >
> > I have set two raid0 arrays on two hardrives, using 2x2 partitions. I did
> > a benchmark of the resulting arrays and one is much slower than the other
> > one. I used bonnie++ and hdparm for the tests. It showed that /dev/md0 is
> > 30% slower than /dev/md1.
> > /dev/md1 is really close to twice the performance of a single drive.
> > I set the arrays with the same parameters. md0 is built from 2x 5 Gb
> > while md1 is built from 2x2 Gb. The harddrives have the same partition
> > table. md0 is the closest to the begining of the drives.
> > The filessystem is Reiserfs on both arrays.
> > Everything works well, except i am curious about such a difference in
> > performances, since the arrays share the same hardware.
> > I am new to linux raid, i do not know if this is normal or weird.
>
> This is to be expected.  Partitions toward the "end" of the disk are
> written closer to the center of the platters, and as a result, data
> transfers suffer.  The drive spins at a constant number of revolutions
> per minute, making the linear velocity of the platter across the read
> head much higher at the outside of the disk than the outside.  HTH,
>       Greg

Re: Performance difference between two raid0 arrays on same drives?

[Gordon Henderson]

On Sat, 12 Jul 2003, Michel Bellais wrote:

> You're right, I thought about it too, but the fastest array is built with
> partitions closer to the centre of the disk, so it should be the slowest
> indeed.
> The disks are big (180 Gb), the partitions represent less than 10% of it and
> follow each others.  It cannot explain 30% difference in performance.
>
> I have created a third array on the disk, which is a copy of the slowest
> array. It has the same content. This last array shows much better performance
> than the original. And it is even closer to the centre...
> So i really don't understand.

Just a thought: What if modern disk manufacturers write from the inside
out, rather than traditionally from the outside in?

CD's are read from the inside out so we can have different size discs.

Or maybe the hard disk manufacturers cottoned on to the fact that most
people would benchmark freshly partitioned disks hoping the file would be
at the "start" so they make it on the inside and get a better benchmark?

Who knows! And I guess without taking one to bits to watch it work it's
going to be hard to find out...

Gordon

Re: Performance difference between two raid0 arrays on same drives?

[Mads Peter Bach]

Michel Bellais wrote:

> You're right, I thought about it too, but the fastest array is built with 
> partitions closer to the centre of the disk, so it should be the slowest 
> indeed.
> The disks are big (180 Gb), the partitions represent less than 10% of it and 
> follow each others.  It cannot explain 30% difference in performance.

You can test the performance across your drives with the program zcav 
(which should come with bonnie++ - if not, take a look at 
http://www.coker.com.au/bonnie++/)

-- 
Mads Peter Bach
Systemadministrator,  Det Humanistiske Fakultet, Aalborg Universitet
Kroghstræde 3 - 5.111, DK-9220 Aalborg Øst - (+45) 96358062
# whois MPB1-DK@whois.dk-hostmaster.dk

-
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: Performance difference between two raid0 arrays on same drives?

[Michel Bellais]

it sounds tricky to me! 
But if some manufacturer inverts the partition order, that would be good for 
linux: usually windows is installed first and takes the best place on the 
drive, the penguin coming second.

On Saturday 12 July 2003 05:20 pm, Gordon Henderson wrote:
> On Sat, 12 Jul 2003, Michel Bellais wrote:
> > You're right, I thought about it too, but the fastest array is built with
> > partitions closer to the centre of the disk, so it should be the slowest
> > indeed.
> > The disks are big (180 Gb), the partitions represent less than 10% of it
> > and follow each others.  It cannot explain 30% difference in performance.
> >
> > I have created a third array on the disk, which is a copy of the slowest
> > array. It has the same content. This last array shows much better
> > performance than the original. And it is even closer to the centre...
> > So i really don't understand.
>
> Just a thought: What if modern disk manufacturers write from the inside
> out, rather than traditionally from the outside in?
>
> CD's are read from the inside out so we can have different size discs.
>
> Or maybe the hard disk manufacturers cottoned on to the fact that most
> people would benchmark freshly partitioned disks hoping the file would be
> at the "start" so they make it on the inside and get a better benchmark?
>
> Who knows! And I guess without taking one to bits to watch it work it's
> going to be hard to find out...
>
> Gordon
>
> -
> To unsubscribe from this list: send the line "unsubscribe linux-raid" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: Performance difference between two raid0 arrays on same drives?

[Michel Bellais]

I have done it with zcat!
The transfer rate decreases steadily from the beginning of the disk to the 
end. It doesn't look linear, but more like a square root..
The loss is about 50% at the end of the disk compared to the beginning.

The test shows that the decrease in performance is negligible in the first 10% 
of the disk, where the raid arrrays are.
So i have really no clue about this performance loss.

Thanks for the tip! It was interesting to test this drive.
Michel Bellais

>
> You can test the performance across your drives with the program zcav
> (which should come with bonnie++ - if not, take a look at
> http://www.coker.com.au/bonnie++/)
>
>--
>Mads Peter Bach
>Systemadministrator,  Det Humanistiske Fakultet, Aalborg Universitet
>Kroghstræde 3 - 5.111, DK-9220 Aalborg Øst - (+45) 96358062
># whois MPB1-DK@whois.dk-hostmaster.dk

-
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

A simple question

[Donghui Wen]

Hi,
   I have a simple question:
    Can software raid survive one disk hot unplug when the raid is under
heavy load?

    During my test (3ware 7500-8 jbod mode, software raid 5), I ran bonnie++
on the
top of md and then unplugged one disk. I saw a lot of scsi disk errors from
dmesg, and then
scsi layer marked this disk offline. Finnally md detected disk failure and
mark this disk
as faulty. But bonnie++ still hangs there with state 'D' which could not be
killed by -9.

Thanks!

Donghui Wen

Re: Performance difference between two raid0 arrays on same drives?

[Gregory Leblanc]

[-- Attachment #1: Type: text/plain, Size: 1211 bytes --]

On Fri, 2003-07-11 at 09:13, Michel wrote:
> Hi!
> 
> I have set two raid0 arrays on two hardrives, using 2x2 partitions. I did a 
> benchmark of the resulting arrays and one is much slower than the other one. 
> I used bonnie++ and hdparm for the tests. It showed that /dev/md0 is 30% 
> slower than /dev/md1. 
> /dev/md1 is really close to twice the performance of a single drive.
> I set the arrays with the same parameters. md0 is built from 2x 5 Gb while md1 
> is built from 2x2 Gb. The harddrives have the same partition table. md0 is 
> the closest to the begining of the drives.
> The filessystem is Reiserfs on both arrays.
> Everything works well, except i am curious about such a difference in 
> performances, since the arrays share the same hardware. 
> I am new to linux raid, i do not know if this is normal or weird.

This is to be expected.  Partitions toward the "end" of the disk are
written closer to the center of the platters, and as a result, data
transfers suffer.  The drive spins at a constant number of revolutions
per minute, making the linear velocity of the platter across the read
head much higher at the outside of the disk than the outside.  HTH,
	Greg


[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Performance difference between two raid0 arrays on same drives?

[Michel]

Hi!

I have set two raid0 arrays on two hardrives, using 2x2 partitions. I did a 
benchmark of the resulting arrays and one is much slower than the other one. 
I used bonnie++ and hdparm for the tests. It showed that /dev/md0 is 30% 
slower than /dev/md1. 
/dev/md1 is really close to twice the performance of a single drive.
I set the arrays with the same parameters. md0 is built from 2x 5 Gb while md1 
is built from 2x2 Gb. The harddrives have the same partition table. md0 is 
the closest to the begining of the drives.
The filessystem is Reiserfs on both arrays.
Everything works well, except i am curious about such a difference in 
performances, since the arrays share the same hardware. 
I am new to linux raid, i do not know if this is normal or weird.
Thanks for any help!

/etc/raidtab:
raiddev /dev/md0
	raid-level 0
	nr-raid-disks 2
	persistent-superblock 1
	chunk-size 32
	device /dev/hde3
	raid-disk 0
	device /dev/hdg3
	raid-disk 1

raiddev /dev/md1
	raid-level 0
	nr-raid-disks 2
	persistent-superblock 1
	chunk-size 32
	device /dev/hde5
	raid-disk 0
	device /dev/hdg5
	raid-disk 1