On Wednesday 11 June 2008 07:49, *vertigo* wrote
in /comp.os.linux.hardware:/

> Hello
>
> I have mainboard: gigabyte P965S3.
> I have 2 SATA interfaces using builtin Jmicron controller
> (and another 4 SATA interfaces)
> I upgraded bios to version F12.
> I want to make mirror raid.
>
> What's better to use: software or hardware(jmicron)?

Hardware RAID is always preferable over software RAID *if* it is available.
However, I sincerely doubt that the onboard SATA RAID of your motherboard
will be a hardware RAID.

Most likely, it'll be a hardware-assisted software RAID, or if you will, a
"WinRAID", just like a Winmodem isn't a real modem.

> 1. Using hardware, jmicron (creating raid from bios).
> Will gentoo see it ? Is it stable ?

I do not know the JMicron, but true hardware SATA RAID from a chip on the
motherboard is extremely rare and would also make the board more expensive
than its peers.

Even if you enable the RAID functionality of an SATA onboard
hardware-assisted software RAID, this will only be seen as a RAID by
Windows, and then will require you to load the appropriate driver for it,
if Windows doesn't have a driver for it by itself.

In GNU/Linux, such a device is seen as a simple SATA controller, regardless
of whether the RAID functionality is enabled or not, because basically
that's what it is.

> If one disk fails how will i know ?

Both true hardware RAID and software RAID would report this via /syslogd,/
or possibly via an additional daemon like Adaptec's Storage Manager for
Adaptec RAID adapters.

> Will there be no problem with replacing bad disk with new one ?

The replacement of a failed drive in a RAID array is always subject to
certain delays and restrictions. Not all RAID adapters (or hard disks!)
support hotplugging.

When a drive has been replaced, most hardware RAID adapters will
automatically rebuild the array - which takes some time and will slow down
the machine's performance somewhat - but in software RAID, I believe you
have to rebuild the array manually using /mdadm./

--
*Aragorn*
(registered GNU/Linux user #223157)

>
>> Hello
>>
>> I have mainboard: gigabyte P965S3.
>> I have 2 SATA interfaces using builtin Jmicron controller
>> (and another 4 SATA interfaces)
>> I upgraded bios to version F12.
>> I want to make mirror raid.
>>
>> What's better to use: software or hardware(jmicron)?
>
> Hardware RAID is always preferable over software RAID *if* it is
> available.
> However, I sincerely doubt that the onboard SATA RAID of your motherboard
> will be a hardware RAID.
>
> Most likely, it'll be a hardware-assisted software RAID, or if you will,
> a
> "WinRAID", just like a Winmodem isn't a real modem.
>
>> 1. Using hardware, jmicron (creating raid from bios).
>> Will gentoo see it ? Is it stable ?
>
> I do not know the JMicron, but true hardware SATA RAID from a chip on the
> motherboard is extremely rare and would also make the board more
> expensive
> than its peers.
>
> Even if you enable the RAID functionality of an SATA onboard
> hardware-assisted software RAID, this will only be seen as a RAID by
> Windows, and then will require you to load the appropriate driver for it,
> if Windows doesn't have a driver for it by itself.
>
> In GNU/Linux, such a device is seen as a simple SATA controller,
> regardless
> of whether the RAID functionality is enabled or not, because basically
> that's what it is.
>
>> If one disk fails how will i know ?
>
> Both true hardware RAID and software RAID would report this via
> /syslogd,/
> or possibly via an additional daemon like Adaptec's Storage Manager for
> Adaptec RAID adapters.
>
>> Will there be no problem with replacing bad disk with new one ?
>
> The replacement of a failed drive in a RAID array is always subject to
> certain delays and restrictions. Not all RAID adapters (or hard disks!)
> support hotplugging.
>
> When a drive has been replaced, most hardware RAID adapters will
> automatically rebuild the array - which takes some time and will slow
> down
> the machine's performance somewhat - but in software RAID, I believe you
> have to rebuild the array manually using /mdadm./
>


thanx for this info. That's true - jmicron is not real hardware RAID.
Linux kernel has drivers for jmicron software/hardware RAID.

So - what is better ?
1. fully software RAID based on mdadm tool
2. or software/hardware jmicron RAID ?

What would you suggest ?
What is faster ? What makes CPU works less ?
Which solution is more reliable ?

Thanx

On Wednesday 11 June 2008 12:16, *vertigo* wrote
in /comp.os.linux.hardware:/

>>> [...]
>>> 1. Using hardware, jmicron (creating raid from bios).
>>> Will gentoo see it ? Is it stable ?
>>
>> I do not know the JMicron, but true hardware SATA RAID from a chip on the
>> motherboard is extremely rare and would also make the board more
>> expensive than its peers.
>>
>> Even if you enable the RAID functionality of an SATA onboard
>> hardware-assisted software RAID, this will only be seen as a RAID by
>> Windows, and then will require you to load the appropriate driver for it,
>> if Windows doesn't have a driver for it by itself.
>>
>> In GNU/Linux, such a device is seen as a simple SATA controller,
>> regardless of whether the RAID functionality is enabled or not, because
>> basically that's what it is.
>>
>> [...]
>
> thanx for this info. That's true - jmicron is not real hardware RAID.
> Linux kernel has drivers for jmicron software/hardware RAID.
>
> So - what is better ?
> 1. fully software RAID based on mdadm tool
> 2. or software/hardware jmicron RAID ?

In this particular case, there wouldn't be much of a difference
performancewise. However, if the Linux kernel has a driver for this chip,
then it might be wise to use it, just in the event that using the chipset
with the proper driver provides extra functionality - e.g. with regard to
error or status reporting - over the default software RAID mechanisms in
the rather "universal" Linux software RAID.

> What would you suggest ?

Try using the /jmicron/ driver and see what gives. If stability issues were
to ensue from this driver, then you know you have a safe fallback solution
using traditional Linux software RAID.

> What is faster ? What makes CPU works less ?

There would normally not be any difference in performance or CPU load
between a complete software RAID and a hardware-assisted software RAID.
They both use the mechanism of software RAID, but the "native" driver might
add some functionality.

> Which solution is more reliable ?

Well, I have no experience with or reports on the /jmicron/ driver, but what
I do know is that Linux software RAID is a timeproven and robust solution.

If however the /jmicron/ driver is part of the stock Linux kernel tree, then
it means that the driver is released under the GPL, and if so, it'll be
quite reliable as well, because Linus would never sanction the inclusion of
an unstable driver in his stable kernel tree.

Just as an illustration (and for what it's worth), a survey conducted by an
American university - I forgot which one - a few years ago showed that if
one compares two equal amounts of proprietary and FOSS code - i.e. an equal
number of lines of code - then the proprietary code contains about 300% to
400% as many bugs as the comparable amount of FOSS code.

So in other words, if the driver is included in the stock Linux kernel, then
it should statistically be considered about 300 to 400 times more stable
than a proprietary driver, even if that proprietary driver were released
from the hardware manufacturers themselves - cfr. nVidia and the likes.

Bottom line: If this /jmicron/ driver is indeed part of the stock kernel - I
keep repeating this because in all honesty I don't know whether it is; I
myself tend to favor SCSI solutions instead for RAID and I'm therefore less
interested in the PATA/SATA RAID stuff - then the reliability of this
driver /should/ be on par with the generic Linux software RAID mechanisms.

Basically, the only difference between regular software RAID and
the /jmicron/ driver would be the driver for the chipset. In the former
case, you would be using a generic SATA driver, and in the latter a driver
that appeals to the extra functionality in this chip.

Hope this was helpful... ;-)

--
*Aragorn*
(registered GNU/Linux user #223157)

vertigo writes:
>What's better to use: software or hardware(jmicron)?
>
>1. Using hardware, jmicron (creating raid from bios).
>Will gentoo see it ? Is it stable ?
>If one disk fails how will i know ?
>Will there be no problem with replacing bad disk with new one ?
>
>2. Software raid (using mdadm)
>
>What's better ? Has anybody used this jmicron raid ?

"Hardware" RAID has the following disadvantages:

- When the controller dies, you to install the same hardware to get at
your data. At that time, you will probably not be able to buy it
anymore, so better buy it now. If you do RAID1, then you may be
able to get to the data without additional hardware (but better try
that first to be sure), but then you won't have a performance
advantage over md even if the jmicron RAID is really hardware RAID.

- "Hardware" RAID works on whole disks, md can be applied to
partitions; so with md you can RAID1 your valuable data, RAID0 some
scratch stuff, and have non-RAID swap partitions.

In conclusion, I recommend using md unless you have really good
reasons to use something else.

BTW, I just wasted a day playing around with an ICH9R-based RAID; the
idea was to use the ICH9R RAID1 on Windows and md on Linux, but it
does not work that way: contrary to what some people have claimed,
Intel's Matrix-RAID does not work on partitions, but allows to divide
the disks into two (not more) parts in a different way; the idea is
probably that you use RAID1 for one part and RAID0 for the other part.
Anyway, I finally gave up on that idea, so Windows just has to do
without RAID; it's not so important anyway:-).

The aftereffects of my experiments with the RAID cost me another
half-day (even after changing the mode to AHCI and writing over the
start of the disk, the Windows driver still saw the RAID I had created
during the experiments.

One other thing worth noting: For md, make sure you create RAIDs with
a version 0.90 superblock (mdadm option -e 0.90), because lilo (if you
boot from RAID) and kernels as recent as 2.6.19 don't grok the newer
superblock formats; even though the documentation says that 0.90 is
the default, the Knoppix (5.3 IIRC) mdadm created 1.0 superblocks for
me by default:-(.

- anton
--
M. Anton Ertl Some things have to be seen to be believed
anton@mips.complang.tuwien.ac.at Most things have to be believed to be seen
http://www.complang.tuwien.ac.at/anton/home.html

On Wed, 11 Jun 2008 11:33:03 +0200 Aragorn wrote:

| Hardware RAID is always preferable over software RAID *if* it is available.
| However, I sincerely doubt that the onboard SATA RAID of your motherboard
| will be a hardware RAID.

I have a Tyan S2927A2NRF with 6 SATA ports. I had 4 of them connected for
a while to 4 WD 750GB SATA drives. I achieved 96MB/s transfer rate when
accessing 1 drive at a time (linear reading/writing so the drive would be
doing more data transfer than seeking for this test). When I ran all 4
drives at the same time (4 separate processes doing reading), the speed
of each dropped to about 92MB/s, so I was getting an aggregate speed of
368MB/s. I don't know what caused the slowdown (controller, bus, driver,
scheduler) but the machine has 2x dual-core 2.8GHz Opeterons, so there is
at least enough CPU power there to work 4 processes easily. Nevertheless
it's still a decent throughput.

My question is, does your preference for hardware RAID still hold here?
Presumably I could get a 368MB/s peak with software RAID minus whatever
overhead that RAID would incur. Hardware RAID in the controller (it has
it but I've not determined if Linux can configure and work with it, and
it might be the "WinRAID" you mention) might do just as well. External
hardware RAID might be a problem as the aggregate transfer rate (2.944
Gbps) is pushing the limit of a SATA-II connection (hypothetically 3.0
Gbps).


| Most likely, it'll be a hardware-assisted software RAID, or if you will, a
| "WinRAID", just like a Winmodem isn't a real modem.
[*] El-cheapo devices that steal resources from the main CPU(s) so they can
market lower pricing.


|> 1. Using hardware, jmicron (creating raid from bios).
|> Will gentoo see it ? Is it stable ?
|
| I do not know the JMicron, but true hardware SATA RAID from a chip on the
| motherboard is extremely rare and would also make the board more expensive
| than its peers.

Seems to be common on server grade motherboards in the many-hundred-dollar
ranges.


| Even if you enable the RAID functionality of an SATA onboard
| hardware-assisted software RAID, this will only be seen as a RAID by
| Windows, and then will require you to load the appropriate driver for it,
| if Windows doesn't have a driver for it by itself.
|
| In GNU/Linux, such a device is seen as a simple SATA controller, regardless
| of whether the RAID functionality is enabled or not, because basically
| that's what it is.

I'm guessing if it is a "WinRAID" style, it has special queueing commands to
at least improve up having N separate SATA ports doing as software RAID.


|> If one disk fails how will i know ?
|
| Both true hardware RAID and software RAID would report this via /syslogd,/
| or possibly via an additional daemon like Adaptec's Storage Manager for
| Adaptec RAID adapters.
|
|> Will there be no problem with replacing bad disk with new one ?
|
| The replacement of a failed drive in a RAID array is always subject to
| certain delays and restrictions. Not all RAID adapters (or hard disks!)
| support hotplugging.

It appears the ones in the Tyan S2927A2NRF board do not. I connected a
couple ports to an eSATA adapter and tried plugging in an external SATA
drive (A WD 1TB drive with 3 interfaces). It would not recognize the
hotplugging automatically. If I had the drive plugged in while booting
up, it would find the drive OK.


| When a drive has been replaced, most hardware RAID adapters will
| automatically rebuild the array - which takes some time and will slow down
| the machine's performance somewhat - but in software RAID, I believe you
| have to rebuild the array manually using /mdadm./

Presumably some kind of RAID manager daemon could run in the background to
emulate the hardware RAID by doing this chore when it sees the drive being
added.

For a project I am studying which will need a lot of reliable file space,
I'm looking at using RAID only to expand the storage per server, and doing
redundancy between servers, with load balancing between servers as well.
Programs would run on the servers to compare each server tree to the others
and replicate anything found missing on one or more servers.

--
|WARNING: Due to extreme spam, googlegroups.com is blocked. Due to ignorance |
| by the abuse department, bellsouth.net is blocked. If you post to |
| Usenet from these places, find another Usenet provider ASAP. |
| Phil Howard KA9WGN (email for humans: first name in lower case at ipal.net) |

On Saturday 14 June 2008 22:19, someone who identifies as
*phil-news-nospam@ipal.net* wrote in /comp.os.linux.hardware:/

> On Wed, 11 Jun 2008 11:33:03 +0200 Aragorn
> wrote:
>
> | Hardware RAID is always preferable over software RAID *if* it is
> | available. However, I sincerely doubt that the onboard SATA RAID of your
> | motherboard will be a hardware RAID.
>
> I have a Tyan S2927A2NRF with 6 SATA ports. I had 4 of them connected for
> a while to 4 WD 750GB SATA drives. I achieved 96MB/s transfer rate when
> accessing 1 drive at a time (linear reading/writing so the drive would be
> doing more data transfer than seeking for this test). When I ran all 4
> drives at the same time (4 separate processes doing reading), the speed
> of each dropped to about 92MB/s, so I was getting an aggregate speed of
> 368MB/s. I don't know what caused the slowdown (controller, bus, driver,
> scheduler) but the machine has 2x dual-core 2.8GHz Opeterons, so there is
> at least enough CPU power there to work 4 processes easily. Nevertheless
> it's still a decent throughput.

So it is, and as long as your disks don't have a higher throughput than the
bus can handle, you're okay. ;-)

> My question is, does your preference for hardware RAID still hold here?

Well, my preference for hardware RAID is not based upon performance alone,
but also on other technicalities.

> Presumably I could get a 368MB/s peak with software RAID minus whatever
> overhead that RAID would incur. Hardware RAID in the controller (it has
> it but I've not determined if Linux can configure and work with it, and
> it might be the "WinRAID" you mention) might do just as well. External
> hardware RAID might be a problem as the aggregate transfer rate (2.944
> Gbps) is pushing the limit of a SATA-II connection (hypothetically 3.0
> Gbps).

I have no experience with real SATA hardware RAID, but I do have an Adaptec
SAS hardware RAID controller in my Tyan machine, and it has a 3 Gb/sec (384
MB/sec) bus speed - and this is important! - *per* attached storage device.

In other words, the transfer rate of the connection between the controller
and each individual disk is 384 MB/sec. The only possible bottleneck -
should the disks really be able to put out that much data per second, which
they don't - is the connection between the motherboard and the SAS
controller, but in this case that's an 8-lane PCIe connection, so I'm safe
there. ;-)

> |> 1. Using hardware, jmicron (creating raid from bios).
> |> Will gentoo see it ? Is it stable ?
> |
> | I do not know the JMicron, but true hardware SATA RAID from a chip on
> | the motherboard is extremely rare and would also make the board more
> | expensive than its peers.
>
> Seems to be common on server grade motherboards in the many-hundred-dollar
> ranges.

Well, I believe my Tyan n6650W motherboard was around 1400 Euro - I'm not
sure, I'd have to check - and it has a single channel onboard IDE
controller, an onboard SATA controller with /nvraid/ - i.e. a
hardware-assisted software RAID from the nVidia nForce Professional chipset
- plus an onboard non-RAID LSI SAS controller and an onboard Firewire
controller.

> | Even if you enable the RAID functionality of an SATA onboard
> | hardware-assisted software RAID, this will only be seen as a RAID by
> | Windows, and then will require you to load the appropriate driver for
> | it, if Windows doesn't have a driver for it by itself.
> |
> | In GNU/Linux, such a device is seen as a simple SATA controller,
> | regardless of whether the RAID functionality is enabled or not, because
> | basically that's what it is.
>
> I'm guessing if it is a "WinRAID" style, it has special queueing commands
> to at least improve up having N separate SATA ports doing as software
> RAID.

I'm not really well-versed on such devices, but you might be correct about
that.

> | When a drive has been replaced, most hardware RAID adapters will
> | automatically rebuild the array - which takes some time and will slow
> | down the machine's performance somewhat - but in software RAID, I
> | believe you have to rebuild the array manually using /mdadm./
>
> Presumably some kind of RAID manager daemon could run in the background to
> emulate the hardware RAID by doing this chore when it sees the drive being
> added.

Something like that would have to be handled via /udevd,/ I presume.

> For a project I am studying which will need a lot of reliable file space,
> I'm looking at using RAID only to expand the storage per server, and doing
> redundancy between servers, with load balancing between servers as well.
> Programs would run on the servers to compare each server tree to the
> others and replicate anything found missing on one or more servers.

The easiest and probably most reliable solution would be to use hardware
RAID controllers, then, in my humble opinion. But I presume you're going
to hook up SATA drives to them? I can't really say that I have much
experience with those as I've been using SCSI drives for many years
now. :-/

--
*Aragorn*
(registered GNU/Linux user #223157)

vertigo wrote:
> Hello
>
> I have mainboard: gigabyte P965S3.
> I have 2 SATA interfaces using builtin Jmicron controller
> (and another 4 SATA interfaces)
> I upgraded bios to version F12.
> I want to make mirror raid.
>
> What's better to use: software or hardware(jmicron)?

Software RAID. There are several failure modes that you can experience
-- drive failure (probably the most common), or controller failure (and
some others, but we won't get into it).

If a drive fails, software and hardware RAID are equivalent. Simply
replace the drive. The md software in Linux will automatically rebuild
the drive as needed. The "hardware" solution? Who knows, but we presume
it will do the same.

The real difference is what happens if you suffer a controller failure.
Typically, if using hardware RAID (say RAID 5), you will have to replace
the mainboard/controller with an EXACT match. Which puts a lot of trust
in your vendor. With Software RAID, you put the drives into a compatible
Linux box, and things "just work".

I would stay away from Gentoo for this application, and stick to a
(larger) distribution (say SuSE). If there is a controller problem, it
is very likely that the disks can then be moved to a new
mainboard/controller and even BOOT there (no installation required,
faster system recovery). Also, the application is purely i/o bound, and
DMA is used to move the data anyway. Computing RAID 5 crcs may seem to
be an issue, but any modern processor can do this very quickly (it is
not a bottleneck anymore, even if done in software).

I use mirroring for the boot partition, and RAID 5 for the data
partitions. Linux cannot boot from a RAID 5 volume, but can boot from a
mirrored volume. I also tend to put 4 drives into a RAID 5 volume. This
gives 3/4 of the available raw storage for data (4 500 GB drives gives
1.5TB of available space, with a ability to lose any single drive and
continue).

>
> 1. Using hardware, jmicron (creating raid from bios).
> Will gentoo see it ? Is it stable ?
> If one disk fails how will i know ?
> Will there be no problem with replacing bad disk with new one ?
>
> 2. Software raid (using mdadm)
>
> What's better ? Has anybody used this jmicron raid ?
>
> thanx

Fred Weigel wrote:
> vertigo wrote:
>> Hello
>>
>> I have mainboard: gigabyte P965S3.
>> I have 2 SATA interfaces using builtin Jmicron controller
>> (and another 4 SATA interfaces)
>> I upgraded bios to version F12.
>> I want to make mirror raid.
>>
>> What's better to use: software or hardware(jmicron)?
>
> Software RAID. There are several failure modes that you can experience
> -- drive failure (probably the most common), or controller failure (and
> some others, but we won't get into it).
>
Personally, I've generally preferred hardware RAID. I've had software
RAID systems not notice a drive failing, causing corruption of the
filesystem. The better hardware RAID cards seem to be better at sniffing
out a failing disk and removing it from an array before any problems can
ensue.

> If a drive fails, software and hardware RAID are equivalent. Simply
> replace the drive. The md software in Linux will automatically rebuild
> the drive as needed. The "hardware" solution? Who knows, but we presume
> it will do the same.
>
All hardware RAID systems I've used do this and more. They'll even
e-mail you with any alerts on the system.

> The real difference is what happens if you suffer a controller failure.
> Typically, if using hardware RAID (say RAID 5), you will have to replace
> the mainboard/controller with an EXACT match. Which puts a lot of trust
> in your vendor. With Software RAID, you put the drives into a compatible
> Linux box, and things "just work".
>
I wouldn't have thought this is such a major problem if you use a major
RAID system vendor. 3Ware or Areca are likely to be around for a while.
For really important data systems, you should be thinking of having
spares ready to be installed, or go for a proper external RAID system
with hot-swap everything and support contracts to match.

If you intend using a large number of disks, you'll need to be using one
of these cards to get the requisite number of ports anyway, even if you
choose to use them as JBOD suppliers to an md RAID.

Matthew

Matthew Wild writes:
>Anton Ertl wrote:
>> Matthew Wild writes:
>>> Personally, I've generally preferred hardware RAID. I've had software
>>> RAID systems not notice a drive failing,
>>
>> Which software RAID was that?
>>
>Standard Linux md RAID.
>
>>> causing corruption of the
>>> filesystem.
>>
>> How does not noticing a failing drive cause the corruption of the file
>> system?
>>
>Because as one drive is failing it is corrupting any accesses to that disk.
....
>I was just pointing out that while a disk is failing, the software RAID
>has not noticed and limped along still trying to use the failing disk
>which happily provides garbage when accessed.

It's an unusual failure mode if a drive delivers wrong data without
reporting an error. Why do you believe that the hardware RAID would
have fared better when confronted with such a drive?

>In my experience, the
>hardware RAID systems I have used, 3Ware, Digital/Compaq/HP RA8000,
>MA8000, MSA1500 (quite horrible to administer), manage their arrays
>fairly conservatively and drop disks pretty quickly.

In my experience md is overly conservative and drops drives pretty
quickly: We had a box that now and then reported IDE errors that the
kernel could recover from by doing an IDE reset; however, by then md
had assumed that the drive had failed and had dropped it from the
array, so we needed to re-add it to the array manually, which was
pretty annoying. OTOH, maybe this annoyance was good, because we
eventually fixed the problem.

- anton
--
M. Anton Ertl Some things have to be seen to be believed
anton@mips.complang.tuwien.ac.at Most things have to be believed to be seen
http://www.complang.tuwien.ac.at/anton/home.html