What is "best" on Linux 2.4.X (with any appropriate patches) ?
HPT370 RAID-1 or Software RAID-1 ?
And if Software RAID-1 is best of these two, where is it most stable; on an
i815 chipset, a VIA chipset (686B) or on a promise controller?
I define "best" as:
* most stable
* least complex setup
* least hassle when something goes wrong
_____________________________________________________
| Martin Eriksson <[email protected]>
| MSc CSE student, department of Computing Science
| Ume? University, Sweden
Martin Eriksson wrote:
> What is "best" on Linux 2.4.X (with any appropriate patches) ?
>
> HPT370 RAID-1 or Software RAID-1 ?
>
> And if Software RAID-1 is best of these two, where is it most stable; on an
> i815 chipset, a VIA chipset (686B) or on a promise controller?
>
> I define "best" as:
> * most stable
> * least complex setup
> * least hassle when something goes wrong
Software RAID is just your disk configuration. But I'd recommend
hardware raid because the rebuild after one disk crash is dog slow with
software raid. This problem been discussed in all possible linux
magazines...
I prefer SCSI RAID5
Thunder
--
begin-base64 755 -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====
Extract this and see what will happen if you execute my
signature. Just save it to file and do a
> uudecode $file | perl
> Software RAID is just your disk configuration. But I'd recommend
> hardware raid because the rebuild after one disk crash is dog slow with
> software raid. This problem been discussed in all possible linux
> magazines...
The raid rebuild time is identical for pretty much any set up. With the
softraid its intentionally defaulting to a low fraction of I/O bandwidth
so it doesnt disrupt normal operation.
Also as far is his question goes - both are software raid
HPT370 IDE Raid is not really an hardware raid.
It is a software Raid, since Linux does not use tha raid implementation
that comes with the BIOS, but it uses softwareraid.
On Thu, 14 Mar 2002, Thunder from the hill wrote:
> Martin Eriksson wrote:
>
> > What is "best" on Linux 2.4.X (with any appropriate patches) ?
> >
> > HPT370 RAID-1 or Software RAID-1 ?
> >
> > And if Software RAID-1 is best of these two, where is it most stable; on an
> > i815 chipset, a VIA chipset (686B) or on a promise controller?
> >
> > I define "best" as:
> > * most stable
> > * least complex setup
> > * least hassle when something goes wrong
>
> Software RAID is just your disk configuration. But I'd recommend
> hardware raid because the rebuild after one disk crash is dog slow with
> software raid. This problem been discussed in all possible linux
> magazines...
> I prefer SCSI RAID5
>
> Thunder
> --
> begin-base64 755 -
> IyEgL3Vzci9iaW4vcGVybApteSAgICAgJHNheWluZyA9CSMgVGhlIHNjcmlw
> dCBvbiB0aGUgbGVmdCBpcyB0aGUgcHJvb2YKIk5lbmEgaXN0IGVpbiIgLgkj
> IHRoYXQgaXQgaXNuJ3QgYWxsIHRoZSB3YXkgaXQgc2VlbXMKIiB2ZXJhbHRl
> dGVyICIgLgkjIHRvIGJlIChlc3BlY2lhbGx5IG5vdCB3aXRoIG1lKQoiTkRX
> LVN0YXIuXG4iICA7CiRzYXlpbmcgPX4Kcy9ORFctU3Rhci9rYW5uXAogdW5z
> IHJldHRlbi9nICA7CiRzYXlpbmcgICAgICAgPX4Kcy92ZXJhbHRldGVyL2Rp
> XAplIExpZWJlL2c7CiRzYXlpbmcgPX5zL2Vpbi8KbnVyL2c7JHNheWluZyA9
> fgpzL2lzdC9zYWd0LC9nICA7CiRzYXlpbmc9fnMvXG4vL2cKO3ByaW50Zigk
> c2F5aW5nKQo7cHJpbnRmKCJcbiIpOwo=
> ====
> Extract this and see what will happen if you execute my
> signature. Just save it to file and do a
> > uudecode $file | perl
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/
>
----- Original Message -----
From: "Luigi Genoni" <[email protected]>
To: "Thunder from the hill" <[email protected]>
Cc: <[email protected]>; "Martin Eriksson" <[email protected]>
Sent: Friday, March 15, 2002 10:31 AM
Subject: Re: HPT370 RAID-1 or Software RAID-1, what's "best"?
> HPT370 IDE Raid is not really an hardware raid.
> It is a software Raid, since Linux does not use tha raid implementation
> that comes with the BIOS, but it uses softwareraid.
Well, even if it uses the BIOS raid, the CPU still has to do all the work.
You can easily spot true hardware raid because of the largish PCI boards,
and the big onboard processors. But this is a budget project... so no true
hardware raid.
I have come to a conclusion now though... as this machine will be used
without a monitor, any RAID arrays will be controlled via a web application.
To do this I need software RAID. Also, from the start it will be running the
following config:
Celeron-II CPU on an i815 board, 256MB RAM
P/M: Boot drive, temporary backup storage for CD-R backup (a few gigs)
P/S: RAID-1 disk #1
S/M: CD-RW drive
S/S: RAID-1 disk #2
In the future, if more storage is needed, there will propably be a three to
four-disk RAID-5 array with some/all disks on a promise controller.
Now this has gone off-topic, so please send replies only to me.
PS. the "mdadm" package is very nice. Among other things it solved the
"array is in use" problem.
* Martin Eriksson <[email protected]> [020315 13:59]:
> Now this has gone off-topic, so please send replies only to me.
CC me please. :)
Gru?
Uli
--
'The box said, 'Requires Windows 95 or better', so i installed Linux - TKK 5
Hi,
Luigi Genoni wrote:
> HPT370 IDE Raid is not really an hardware raid.
> It is a software Raid, since Linux does not use tha raid implementation
> that comes with the BIOS, but it uses softwareraid.
That doesn't prevent me from saying that real hardware raid might be
better. But is the thing you wish to say that there's no difference, or
what?
Alan Cox wrote:
> The raid rebuild time is identical for pretty much any set up. With the
> softraid its intentionally defaulting to a low fraction of I/O bandwidth
> so it doesnt disrupt normal operation.
I experienced it took at about twice the time for a rebuild. I don't
exactly remember the test results, and they aren't available to me until
Monday. If you're interested...
Maybe things have changed a lot since last year, when I did the tests.
> Also as far is his question goes - both are software raid
See above.
Thunder
--
begin-base64 755 -
IyEgL3Vzci9iaW4vcGVybApteSAgICAgJHNheWluZyA9CSMgVGhlIHNjcmlw
dCBvbiB0aGUgbGVmdCBpcyB0aGUgcHJvb2YKIk5lbmEgaXN0IGVpbiIgLgkj
IHRoYXQgaXQgaXNuJ3QgYWxsIHRoZSB3YXkgaXQgc2VlbXMKIiB2ZXJhbHRl
dGVyICIgLgkjIHRvIGJlIChlc3BlY2lhbGx5IG5vdCB3aXRoIG1lKQoiTkRX
LVN0YXIuXG4iICA7CiRzYXlpbmcgPX4Kcy9ORFctU3Rhci9rYW5uXAogdW5z
IHJldHRlbi9nICA7CiRzYXlpbmcgICAgICAgPX4Kcy92ZXJhbHRldGVyL2Rp
XAplIExpZWJlL2c7CiRzYXlpbmcgPX5zL2Vpbi8KbnVyL2c7JHNheWluZyA9
fgpzL2lzdC9zYWd0LC9nICA7CiRzYXlpbmc9fnMvXG4vL2cKO3ByaW50Zigk
c2F5aW5nKQo7cHJpbnRmKCJcbiIpOwo=
====
Extract this and see what will happen if you execute my
signature. Just save it to file and do a
> uudecode $file | perl
> > that comes with the BIOS, but it uses softwareraid.
> That doesn't prevent me from saying that real hardware raid might be
> better. But is the thing you wish to say that there's no difference, or
> what?
The CPU can saturate the I/O bandwidth (if not then go upgrade to a pentium)
At that point its totally disk I/O bound so there will be no difference
On Fri, 15 Mar 2002, Thunder from the hill wrote:
> Hi,
>
> Luigi Genoni wrote:
> > HPT370 IDE Raid is not really an hardware raid.
> > It is a software Raid, since Linux does not use tha raid implementation
> > that comes with the BIOS, but it uses softwareraid.
> That doesn't prevent me from saying that real hardware raid might be
> better. But is the thing you wish to say that there's no difference, or
> what?
Hardware RAID is indeed better, but what you get using HPT370 IDE
controlelr is not hardware raid at all. Just read the code of the driver.
You get a software raid, period.
>
> Alan Cox wrote:
> > The raid rebuild time is identical for pretty much any set up. With the
> > softraid its intentionally defaulting to a low fraction of I/O bandwidth
> > so it doesnt disrupt normal operation.
> I experienced it took at about twice the time for a rebuild. I don't
> exactly remember the test results, and they aren't available to me until
> Monday. If you're interested...
> Maybe things have changed a lot since last year, when I did the tests.
>
> > Also as far is his question goes - both are software raid
> See above.
>
> Thunder
> --
> begin-base64 755 -
> IyEgL3Vzci9iaW4vcGVybApteSAgICAgJHNheWluZyA9CSMgVGhlIHNjcmlw
> dCBvbiB0aGUgbGVmdCBpcyB0aGUgcHJvb2YKIk5lbmEgaXN0IGVpbiIgLgkj
> IHRoYXQgaXQgaXNuJ3QgYWxsIHRoZSB3YXkgaXQgc2VlbXMKIiB2ZXJhbHRl
> dGVyICIgLgkjIHRvIGJlIChlc3BlY2lhbGx5IG5vdCB3aXRoIG1lKQoiTkRX
> LVN0YXIuXG4iICA7CiRzYXlpbmcgPX4Kcy9ORFctU3Rhci9rYW5uXAogdW5z
> IHJldHRlbi9nICA7CiRzYXlpbmcgICAgICAgPX4Kcy92ZXJhbHRldGVyL2Rp
> XAplIExpZWJlL2c7CiRzYXlpbmcgPX5zL2Vpbi8KbnVyL2c7JHNheWluZyA9
> fgpzL2lzdC9zYWd0LC9nICA7CiRzYXlpbmc9fnMvXG4vL2cKO3ByaW50Zigk
> c2F5aW5nKQo7cHJpbnRmKCJcbiIpOwo=
> ====
> Extract this and see what will happen if you execute my
> signature. Just save it to file and do a
> > uudecode $file | perl
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/
>
> Hardware RAID is indeed better, but what you get using HPT370 IDE
> controlelr is not hardware raid at all. Just read the code of the driver.
> You get a software raid, period.
Its not always that simple either.
Software raid on aic7xxx totally blows away the Dell/AMI megaraid card I
have, to the point the megaraid now resides in my testing bucket. The promise
Supertrak 100 (now superceded by the SX6000) is also slower than the
software IDE raid, but does use less CPU in RAID5 mode.
Some hardware raid cards do seem to be winners. The Dell Perc2/QC aacraid
based boards (233Mhz ARM etc) really shift. When I've had the chance to
borrow the disks to test I've seen it running over 100Mbytes/second. It
also supports nice stuff like online reconfiguration of active volumes.
[$$stupid from Dell $$notalot from ebay ;)]
Alan
On Fri, 15 Mar 2002, Alan Cox wrote:
> > Hardware RAID is indeed better, but what you get using HPT370 IDE
> > controlelr is not hardware raid at all. Just read the code of the driver.
> > You get a software raid, period.
>
> Its not always that simple either.
>
> Software raid on aic7xxx totally blows away the Dell/AMI megaraid card I
> have, to the point the megaraid now resides in my testing bucket. The promise
> Supertrak 100 (now superceded by the SX6000) is also slower than the
> software IDE raid, but does use less CPU in RAID5 mode.
yes, I know, but I was talking "generaliter"
>
> Some hardware raid cards do seem to be winners. The Dell Perc2/QC aacraid
> based boards (233Mhz ARM etc) really shift. When I've had the chance to
> borrow the disks to test I've seen it running over 100Mbytes/second. It
> also supports nice stuff like online reconfiguration of active volumes.
> [$$stupid from Dell $$notalot from ebay ;)]
On my CISS Compaq array I get quite similar performances ;)
Luigi
On Fri, Mar 15, 2002 at 05:43:38PM +0000, Alan Cox wrote:
> > Hardware RAID is indeed better, but what you get using HPT370 IDE
> > controlelr is not hardware raid at all. Just read the code of the driver.
> > You get a software raid, period.
>
> Its not always that simple either.
>
> Software raid on aic7xxx totally blows away the Dell/AMI megaraid card I
> have, to the point the megaraid now resides in my testing bucket. The promise
> Supertrak 100 (now superceded by the SX6000) is also slower than the
> software IDE raid, but does use less CPU in RAID5 mode.
>
> Some hardware raid cards do seem to be winners. The Dell Perc2/QC aacraid
> based boards (233Mhz ARM etc) really shift. When I've had the chance to
> borrow the disks to test I've seen it running over 100Mbytes/second. It
> also supports nice stuff like online reconfiguration of active volumes.
> [$$stupid from Dell $$notalot from ebay ;)]
Yep, but the aacraid controllers based on i960 don't do so well. I was able
to double my throughput after switching to software raid (it acts like an
aic7xxx in scsi mode, even using the adaptec driver instead of aacraid).
Mike
On Fri, 15 Mar 2002, Alan Cox wrote:
> > Hardware RAID is indeed better, but what you get using HPT370 IDE
> > controlelr is not hardware raid at all. Just read the code of the driver.
> > You get a software raid, period.
>
> Its not always that simple either.
> Software raid on aic7xxx totally blows away the Dell/AMI megaraid card I
> have, to the point the megaraid now resides in my testing bucket. The promise
> Supertrak 100 (now superceded by the SX6000) is also slower than the
> software IDE raid, but does use less CPU in RAID5 mode.
>
> Some hardware raid cards do seem to be winners. The Dell Perc2/QC aacraid
> based boards (233Mhz ARM etc) really shift. When I've had the chance to
> borrow the disks to test I've seen it running over 100Mbytes/second. It
> also supports nice stuff like online reconfiguration of active volumes.
> [$$stupid from Dell $$notalot from ebay ;)]
Another issue is that not all RAID hardware supports anything below the
drive level, while kernel RAID runs at the partition level, and allows
various things to be in partitions on drives, rather than using the whole
drive for one thing.
I have done things like have four drives split into RAID-5 using one set
of partitions, and RAID-0+1 on another, using different stripe sizes for
the two md sets. And I got there after a lot of trying configs, it reall
worked best that way. Putting overflow swap space on drives with RAID
allows some protection against running out of memory, although you don't
usually want your main swap fighing for space on the elevator.
There probably is a RAID controller which can do this just as well, but
I haven't used it. Versatility is the name of the game for kernel RAID.
--
bill davidsen <[email protected]>
CTO, TMR Associates, Inc
Doing interesting things with little computers since 1979.