* [gentoo-user] RAID
@ 2007-03-10 20:59 mwq
2007-03-10 23:11 ` Dan Farrell
` (2 more replies)
0 siblings, 3 replies; 7+ messages in thread
From: mwq @ 2007-03-10 20:59 UTC (permalink / raw
To: gentoo-user
I have one laic question which may not be directly connected to Gentoo but I think you'll forgive me that.
Imagine such a situation: I have two hard drives but drive A is twice faster when reading and writing then drive B. I want to make RAID 0 using A and B. Why are the stripes sizes on both drives excacly the same? (I've googled and didn't find any information about different spripes sizes) I think that using twice greater stripe on A gives more speed then using equal stripes.
And my question is: where am I doing a mistake?
PS
Sorry for my poor English
--
gentoo-user@gentoo.org mailing list
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [gentoo-user] RAID
2007-03-10 20:59 [gentoo-user] RAID mwq
@ 2007-03-10 23:11 ` Dan Farrell
2007-03-11 9:19 ` Boyd Stephen Smith Jr.
2007-03-13 8:14 ` Nelson, David (ED, PAR&D)
2 siblings, 0 replies; 7+ messages in thread
From: Dan Farrell @ 2007-03-10 23:11 UTC (permalink / raw
To: gentoo-user
On Sat, 10 Mar 2007 21:59:41 +0100
mwq <mwq@o2.pl> wrote:
> I have one laic question which may not be directly connected to
> Gentoo but I think you'll forgive me that. Imagine such a situation:
> I have two hard drives but drive A is twice faster when reading and
> writing then drive B. I want to make RAID 0 using A and B. Why are
> the stripes sizes on both drives excacly the same? (I've googled and
> didn't find any information about different spripes sizes) I think
> that using twice greater stripe on A gives more speed then using
> equal stripes. And my question is: where am I doing a mistake? PS
> Sorry for my poor English
>
Could you raid 2 partitions on A and on on B together? Is A bigger
than B as well? If so you can raid A1, A2, and B together to do this
maybe.
--
gentoo-user@gentoo.org mailing list
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [gentoo-user] RAID
2007-03-10 20:59 [gentoo-user] RAID mwq
2007-03-10 23:11 ` Dan Farrell
@ 2007-03-11 9:19 ` Boyd Stephen Smith Jr.
2007-03-13 8:14 ` Nelson, David (ED, PAR&D)
2 siblings, 0 replies; 7+ messages in thread
From: Boyd Stephen Smith Jr. @ 2007-03-11 9:19 UTC (permalink / raw
To: gentoo-user
[-- Attachment #1: Type: text/plain, Size: 1279 bytes --]
On Saturday 10 March 2007, mwq <mwq@o2.pl> wrote about '[gentoo-user]
RAID':
> Imagine such a situation: I have two
> hard drives but drive A is twice faster when reading and writing then
> drive B. I want to make RAID 0 using A and B. Why are the stripes sizes
> on both drives excacly the same?
If stripe sizes aren't consistent, it's takes hundreds more CPU cycles (and
a separate code path) to determine where to write a block. This
calculation has to be repeated every time a block is written so, it can
easily end up limiting your total throughput by making an I/O operation be
CPU bound.
> I think that using twice
> greater stripe on A gives more speed then using equal stripes.
If A is also twice as big, you can do as Dan Farrell suggested to get some
increased performance. If A is the same size, you'd be sacrificing speed
across the whole raid-ed device to potentially gain some speed in the
sections where you can make A's segments bigger.
--
Boyd Stephen Smith Jr. ,= ,-_-. =.
bss03@volumehost.net ((_/)o o(\_))
ICQ: 514984 YM/AIM: DaTwinkDaddy `-'(. .)`-'
http://iguanasuicide.org/ \_/
New GPG Key! Old key expires 2007-03-25. Upgrade NOW!
[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]
^ permalink raw reply [flat|nested] 7+ messages in thread
* RE: [gentoo-user] RAID
2007-03-10 20:59 [gentoo-user] RAID mwq
2007-03-10 23:11 ` Dan Farrell
2007-03-11 9:19 ` Boyd Stephen Smith Jr.
@ 2007-03-13 8:14 ` Nelson, David (ED, PAR&D)
2007-03-15 0:27 ` Dan Farrell
2 siblings, 1 reply; 7+ messages in thread
From: Nelson, David (ED, PAR&D) @ 2007-03-13 8:14 UTC (permalink / raw
To: gentoo-user
> -----Original Message-----
> From: mwq [mailto:mwq@o2.pl]
> Sent: 10 March 2007 21:00
> To: gentoo-user@lists.gentoo.org
> Subject: [gentoo-user] RAID
>
>
> I have one laic question which may not be directly connected
> to Gentoo but I think you'll forgive me that.
> Imagine such a situation: I have two hard drives but drive A
> is twice faster when reading and writing then drive B. I want
> to make RAID 0 using A and B. Why are the stripes sizes on
> both drives excacly the same? (I've googled and didn't find
> any information about different spripes sizes) I think that
> using twice greater stripe on A gives more speed then using
> equal stripes.
> And my question is: where am I doing a mistake?
> PS
> Sorry for my poor English
>
> --
> gentoo-user@gentoo.org mailing list
>
>
Forgive me the maths approach here, but if drive A reads/writes at speed a, and drive B at speed b.
a = 2*b from what you have said.
In raid 0 drive speed is limited by the slower drive (if I recall correctly) so the speed your raid array would be limited to is .... b + b = 2b = a (from above).
Hence you would be as well not bothering with RAID if drive A is 2x as fast as drive B.
Naturally it's not as clear cut but should be pretty close. RAID 0ing these drives would gain you little in terms of speed while any 2 drive RAID 0 setup increases the chance of failure by 2.
If you just want them to appear as one drive look at something like LVM which can create one volume from both drives, although parts of the volume on drive A would be faster than those parts on drive B.
Just my £0.02, feel free to poke holes in my reasoning ;)
--
djn
I do not represent anyone else in emails I send to this list.
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [gentoo-user] RAID
2007-03-13 8:14 ` Nelson, David (ED, PAR&D)
@ 2007-03-15 0:27 ` Dan Farrell
2007-03-15 0:52 ` kashani
0 siblings, 1 reply; 7+ messages in thread
From: Dan Farrell @ 2007-03-15 0:27 UTC (permalink / raw
To: gentoo-user
On Tue, 13 Mar 2007 08:14:28 -0000
"Nelson, David \(ED, PAR&D\)" <David.Nelson2@astrazeneca.com> wrote:
> > -----Original Message-----
> > From: mwq [mailto:mwq@o2.pl]
> > Sent: 10 March 2007 21:00
> > To: gentoo-user@lists.gentoo.org
> > Subject: [gentoo-user] RAID
> >
> >
> > I have one laic question which may not be directly connected
> > to Gentoo but I think you'll forgive me that.
> > Imagine such a situation: I have two hard drives but drive A
> > is twice faster when reading and writing then drive B. I want
> > to make RAID 0 using A and B. Why are the stripes sizes on
> > both drives excacly the same? (I've googled and didn't find
> > any information about different spripes sizes) I think that
> > using twice greater stripe on A gives more speed then using
> > equal stripes.
> > And my question is: where am I doing a mistake?
> > PS
> > Sorry for my poor English
> >
> > --
> > gentoo-user@gentoo.org mailing list
> >
> >
>
> Forgive me the maths approach here, but if drive A reads/writes at
> speed a, and drive B at speed b.
>
> a = 2*b from what you have said.
> In raid 0 drive speed is limited by the slower drive (if I recall
> correctly) so the speed your raid array would be limited to is .... b
> + b = 2b = a (from above).
>
> Hence you would be as well not bothering with RAID if drive A is 2x
> as fast as drive B.
>
> Naturally it's not as clear cut but should be pretty close. RAID 0ing
> these drives would gain you little in terms of speed while any 2
> drive RAID 0 setup increases the chance of failure by 2.
>
> If you just want them to appear as one drive look at something like
> LVM which can create one volume from both drives, although parts of
> the volume on drive A would be faster than those parts on drive B.
>
> Just my £0.02, feel free to poke holes in my reasoning ;)
well, I'll try a bit perhaps. Given my scheme, we should really be
talking about 3 drives -- well, actualy you generally raid
_partitions_.
A = fast disk, or speed thereof
B = slow disk. or speed thereof
A1 = first partition, size R/3, fast disk
A2 = second partition, size R/3, fast disk
B1 = first partition, size R/3, slow disk
raid-0 is A1 + A2 + B1
now let's say you have a file N to pull from the disk. Naturally, one
third of it will come from each partition. that is, two-thirds of it
comes from A, and A being twice as fast as drive B, the one-third
that's on B gets done in about the same amount of time. so N mb is
copied in max( (2/3*N)/A , (1/3*N)/B ) seconds which should be roughly
the same.
if N was on A alone obviously the speed would be N/A seconds. (N/A) >
(2/3*N)/A.
in reality, though, I think the best performance would probaby involve
just using the fast drive. RAID introduces too much overhead to make
up for itself in this situation I think.
--
gentoo-user@gentoo.org mailing list
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [gentoo-user] RAID
2007-03-15 0:27 ` Dan Farrell
@ 2007-03-15 0:52 ` kashani
2007-03-15 12:32 ` Dan Farrell
0 siblings, 1 reply; 7+ messages in thread
From: kashani @ 2007-03-15 0:52 UTC (permalink / raw
To: gentoo-user
Dan Farrell wrote:
> in reality, though, I think the best performance would probaby involve
> just using the fast drive. RAID introduces too much overhead to make
> up for itself in this situation I think.
I'm betting the act of seeking across the platters on the fast drive for
two separate partitions on the disk makes performance truly awful. The
idea of separate stripe sizes in the original post makes sense for
shifting more I/O to drive A, the fast one, without causing weird things
to happen on the physical disc that the software can't optimize for.
However I think messing with stripe sizes is not something Linux
software raid (or any hardware raid I've dealt with) supports.
kashani
--
gentoo-user@gentoo.org mailing list
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [gentoo-user] RAID
2007-03-15 0:52 ` kashani
@ 2007-03-15 12:32 ` Dan Farrell
0 siblings, 0 replies; 7+ messages in thread
From: Dan Farrell @ 2007-03-15 12:32 UTC (permalink / raw
To: gentoo-user
On Wed, 14 Mar 2007 17:52:26 -0700
kashani <kashani-list@badapple.net> wrote:
> Dan Farrell wrote:
>
> > in reality, though, I think the best performance would probaby
> > involve just using the fast drive. RAID introduces too much
> > overhead to make up for itself in this situation I think.
> I'm betting the act of seeking across the platters on the fast drive
> for two separate partitions on the disk makes performance truly
> awful. The idea of separate stripe sizes in the original post makes
> sense for shifting more I/O to drive A, the fast one, without causing
> weird things to happen on the physical disc that the software can't
> optimize for.
I wouldn't mind seeing test results, but that argument has some weight
to it.
> However I think messing with stripe sizes is not something Linux
> software raid (or any hardware raid I've dealt with) supports.
This is a nice idea, but I haven't been able to find evidence of
supporting it.
> kashani
--
gentoo-user@gentoo.org mailing list
^ permalink raw reply [flat|nested] 7+ messages in thread
end of thread, other threads:[~2007-03-15 12:38 UTC | newest]
Thread overview: 7+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2007-03-10 20:59 [gentoo-user] RAID mwq
2007-03-10 23:11 ` Dan Farrell
2007-03-11 9:19 ` Boyd Stephen Smith Jr.
2007-03-13 8:14 ` Nelson, David (ED, PAR&D)
2007-03-15 0:27 ` Dan Farrell
2007-03-15 0:52 ` kashani
2007-03-15 12:32 ` Dan Farrell
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox