On 07 Dec 2004 23:50:35 GMT, Trevor Hemsley
wrote:

> Ok Guys, my international order for a NetCell SyncRAID SR5103 hardware
> SATA RAID card finally arrived yesterday and I have been playing with it
> since.

Thanks for a very informative post Trevor. Please keep us informed of
any progress.

Hi Trevor,

On Tue, 7 Dec 2004 23:50:35 UTC, "Trevor Hemsley"
wrote:

> I installed the remaining
> two drives and cabled them up, went into the Netcell BIOS and deleted
> the 3 disk array and defined a single 5 drive array which it reports as
> 610,400KB. Same comments apply - with 5 drives, you only get the one
> choice if you tell it to create an array with all 5. I could also have
> done a 3 drive RAID XL array of 305,000KB and a RAID 0 or 1 setup with
> the remaining two.

I guess your numbers should read as 'MB' rather than 'KB' here :-)

And you could get 4-drive Raid-5 resulting in 480 GB as well I guess
...


> I still have a few problems that I need to go through and sort out but
> the card itself seems to be working just fine.
>
> Outstanding probems:
> I tried to run diskio on the RAID array to see what sort of speed I was
> getting. Got me a TRAP 000D in OS2DASD instead.

You may have a problem with the OS/2 low-level disk API's, they
use CHS addressing and are limited to 65535 cylinders ...

This results in a total disk-size limit of 502 GiB (514072 MiB)

When I get to a
> convenient point I'll recreate this and post the TRAP to c.o.o.bugs. I
> can run a similar utility that I wrote and this tells me that I'm
> getting close to 100MB/sec when reading from the start of the array. The
> card is installed in a 32bit, 33MHz PCI slot so that may be the limiting
> factor.
>
> LVM seems to be doing something weird. It allows me to create an
> extended partition on the drive that occupies all 610,400KB.

It seems the controller shows the whole array as a single disk then ?

On most (SCSI) RAID systems I know of, you first define an array by
combining a number of disks (say 5 :-), and then you can actually
define 'logical drives' on the array, each of which will show
up as a separate 'physical disk' to the operating system.

That way you could define a 610 GB array, with two 'logical drives'
that are smaller than the 502 GiB OS2 'per disk' limit.

Perhaps this SATA-RAID does not have that flexibility though ...

>I can
> format it with JFS. I shutdown and restart and the drive is missing. Go
> into LVM again and it's siiting there, defined, with a drive letter
> assigned to it but not 'committed' or whatever it is that makes LVM
> place the drive letter just to the left of where it normally resides.
> Use F3 to quit and tell it to save the changes that I didn't make and
> the drive then appears to OS/2. I then run chkdsk against it and it's
> usable again. Actually it's usuable before running chkdsk but I notice
> that running `df -h` does not list drive I: until I run chkdsk but it is
> visible to everything else that I've tried. Reboot and it all needs to
> be done again. Go into lvm, F3 and save it and it's OK, otherwise it's
> just not there. Doesn't affect the data that's on the drive, just a
> nuisance. Because of this problem I have had to specify
> RESERVEDRIVELETTER=K in config.sys or my CD drive gets assigned the I:
> drive letter and then LVM tries to make me reboot if I just save the
> changes

Some of this might be related to the same 502 GiB disk-size limit.

Some operations will be affected by it, some will not ...

There might be tools that do not even allow the full 502 GiB since
they
use another artificial limit on the number of cylinders (like 32735).

Regards, JvW

--
Jan van Wijk; Author of DFSee: http://www.dfsee.com

On Wed, 8 Dec 2004 16:12:25 UTC in comp.os.os2.setup.misc, "Jan van
Wijk" wrote:

> Hi Trevor,
>
> On Tue, 7 Dec 2004 23:50:35 UTC, "Trevor Hemsley"
> wrote:
>
> > I installed the remaining
> > two drives and cabled them up, went into the Netcell BIOS and deleted
> > the 3 disk array and defined a single 5 drive array which it reports as
> > 610,400KB. Same comments apply - with 5 drives, you only get the one
> > choice if you tell it to create an array with all 5. I could also have
> > done a 3 drive RAID XL array of 305,000KB and a RAID 0 or 1 setup with
> > the remaining two.
>
> I guess your numbers should read as 'MB' rather than 'KB' here :-)

Indeed, in other posts I got it right ;-) Just not used to having that
amount of space available and KB sounded more likely to my tired brain!

> And you could get 4-drive Raid-5 resulting in 480 GB as well I guess

It doesn't seem to do RAID 5. It does RAID XL which seems to be a
proprietary name for RAID 3. They talk about a 4 drive RAID 0 array and
then using a 5th drive as a dedicated parity drive.

>
> > I still have a few problems that I need to go through and sort out but
> > the card itself seems to be working just fine.
> >
> > Outstanding probems:
> > I tried to run diskio on the RAID array to see what sort of speed I was
> > getting. Got me a TRAP 000D in OS2DASD instead.
>
> You may have a problem with the OS/2 low-level disk API's, they
> use CHS addressing and are limited to 65535 cylinders ...
>
> This results in a total disk-size limit of 502 GiB (514072 MiB)

I have a very recent version of DANIS506.ADD that reports this drive
with 38608 cyls, 255 tracks/cyl, 127 sectors/track. I also have all the
> 512GB fixes on from IBM including a copy of OS2DASD.DMD dated

12-10-04 1:27p 40554 54 OS2DASD.DMD

I'm told that my system should have all the IBM fixes on it relating to
512GB disks.

Interestingly, I ran sysbench against it expecting that to cause the
same trap but it didn't. It looks like whatever diskio does that causes
it is something that I removed from the code or fixed when I ported Kai
Uwe Rommel's code over for sysbench use. It also does not do it when I
use my own utility that reads the raw disk in 256KB chunks - but that
one craps out with an i/o error when it hits 137GB (decimal) which is
too coincidental to be an accident but not something that I've looked at
to find out if it's fixable yet. I've never left it running for long
enough to read 137GB before now!

[snip]
> It seems the controller shows the whole array as a single disk then ?
>
> On most (SCSI) RAID systems I know of, you first define an array by
> combining a number of disks (say 5 :-), and then you can actually
> define 'logical drives' on the array, each of which will show
> up as a separate 'physical disk' to the operating system.
>
> That way you could define a 610 GB array, with two 'logical drives'
> that are smaller than the 502 GiB OS2 'per disk' limit.
>
> Perhaps this SATA-RAID does not have that flexibility though ...

My MegaRAID Express 500 is this way too I think. You can define multiple
arrays but each array has to occupy all members of the disks that make
it up - you can't split a disk in half and have one half in one array
and the other in another. The Netcell card is quite restrictive in the
arrays it allows you to define.

1 disk - JBOD as a single disk
2 disks - RAID 0 or RAID 1
3 disks - RAID XL array giving space of 2 disks or split into 2 arrays
of 1 disk and 2 disks (see above)
4 disks - RAID 0 array of all 4. They don't mention it in the doc but I
think you should also be able to set this up as 2 x 2 disk arrays.
5 disks - RAID XL array giving space of 4 disks. Again unmentioned but I
suspect you could also set this up as a 2 drive array and a 3 drive
array.

I did find mention on their web site last night that it should allow up
to 4 separate arrays but the space allowed in the BIOS setup to display
the arrays that are defined only has 2 lines and it doesn't look like it
would scroll. I suspect that it is limited to 2 arrays and the doc is
wrong but I can't be bothered to delete the 5 disk array and try other
combinations.

Given that the Netcell card is UK£125 (€170) for the 5 port SATA
version, I'm not surprised it doesn't have all the options of more
expensive RAID cards. I can't think of another way of getting 600GB RAID
disk space for UK£450 (€650).

[snip stuff about non-sticky drive letter]
> Some of this might be related to the same 502 GiB disk-size limit.

This one I now have a bypass for. I've added

lvm /setnameartition,2,disk_i,disk_i
chkdsk i: /f /c

to startup.cmd. The lvm command tells it to rename the partition to the
same name it had already and this seems to auto-fix the drive letter
without user intervention which will at least let me boot without being
present. The chkdsk seems like a good idea since it's skipped the boot
time 'dirty' check.

--
Trevor Hemsley, Brighton, UK.
Trevor-Hemsley at dsl dot pipex dot com

Hi Trevor,

On Wed, 8 Dec 2004 18:19:15 UTC, "Trevor Hemsley" wrote:

> [snip stuff about non-sticky drive letter]
> > Some of this might be related to the same 502 GiB disk-size limit.
>
> This one I now have a bypass for. I've added
>
> lvm /setnameartition,2,disk_i,disk_i
> chkdsk i: /f /c
>
> to startup.cmd. The lvm command tells it to rename the partition to the
> same name it had already and this seems to auto-fix the drive letter
> without user intervention which will at least let me boot without being
> present.

Interesting.

This is more evidence for my suspicion that the code in OS2LVM.DMD
plus OS2DASD.DMD is more critical in finding 'correctly' defined
LVM partitions that the LVM-engine itself in LVM.DLL as used by
the LVM.EXE program does ...

Both handle assigning driveletters to LVM-volumes, but the methods
in the LVM.EXE seem to be able to succeeed more often ...

>The chkdsk seems like a good idea since it's skipped the boot
> time 'dirty' check.

OK

Regards, JvW

--
Jan van Wijk; Author of DFSee: http://www.dfsee.com

On Tue, 7 Dec 2004 23:50:35 UTC in comp.os.os2.setup.storage, "Trevor
Hemsley" wrote:

> I have not tested any of the RAID recovery capabilities of the card and
> I'm not sure I can be bothered to thoguh I suppose I should, just to
> check that it really does work before I need it to work! I did manage to
> make the 3 drive RAID XL array fail by accidently knocking one of the
> cables off and it has options in the BIOS to recover using the same
> drive. I did tell it to do this but it looked like it was going to take
> a very long time and the array was empty so I stopped it, deleted and
> recreated the entire thing instead.

I managed to test this, again by accident. I managed to knock the cable
off the first disk again and that's all it takes to need to rebuild the
array. There's no option to say "just put it back together again" as
there was with the DPT card and its utilities that I used to use. With
that you could boot from DOS and run the utility and tell it to put the
array back together again without rebuilding it - for use when you knew
that no updates had been made like this case.

So this time I told it to rebuild it. I have no idea how long it took
but it took a very long time. I started the rebuild mid-morning and it
was done by sometime before midnight. I reckon it probably took at least
12 hours to rebuild a failed 160GB drive, maybe as long as 14. It
doesn't start to do the rebuild until you exit the BIOS after you tell
it to start the rebuild and there seems to be no indication of its
progress without going back into it - there may be from the Windows
management utility. There's also no indication that the array is not
optimal until you see the controller banner at reboot - Dani's diskinfo
reports the 'disk' as SMART OK even when it's degraded but I have
emailed Netcell and suggested that it might be a good way to support
non-Windows o/ses. While the rebuild was going, I could use the system
without noticing that it was happening. It did show up on a disk
benchmark - before I started the rebuild and while the rebuild was going
it would read at between 35 and 60MB/sec but once it was complete it
went back up to ~100MB/sec.

It does worry me a bit quite how easy it is to make the array fail.

--
Trevor Hemsley, Brighton, UK.
Trevor-Hemsley at dsl dot pipex dot com

On Tue, 7 Dec 2004 23:50:35 UTC in comp.os.os2.setup.storage, "Trevor Hemsley"
wrote:

> LVM seems to be doing something weird. It allows me to create an
> extended partition on the drive that occupies all 610,400KB. I can
> format it with JFS. I shutdown and restart and the drive is missing. Go
> into LVM again and it's siiting there, defined, with a drive letter
> assigned to it but not 'committed' or whatever it is that makes LVM
> place the drive letter just to the left of where it normally resides.
> Use F3 to quit and tell it to save the changes that I didn't make and
> the drive then appears to OS/2. I then run chkdsk against it and it's
> usable again.

Following up on my ~2.5 year old post. It turns out that this weird problem with
my 596GB array not appearing as a drive letter were down to a problem in
DaniS506.ADD which she just fixed. She sent me a copy of 1.7.10p and I can now
reboot and have my array instantly accessible. Wasn't a problem in LVM after all
despite the symptoms appearing to point that way.

--
Trevor Hemsley, Brighton, UK
Trevor dot Hemsley at ntlworld dot com

Trevor Hemsley schrieb:
> On Tue, 7 Dec 2004 23:50:35 UTC in comp.os.os2.setup.storage, "Trevor Hemsley"
> wrote:
>
>> [....]
>
> Following up on my ~2.5 year old post. It turns out that this weird problem with
> my 596GB array not appearing as a drive letter were down to a problem in
> DaniS506.ADD which she just fixed. She sent me a copy of 1.7.10p and I can now
> reboot and have my array instantly accessible. Wasn't a problem in LVM after all
> despite the symptoms appearing to point that way.
>
Good news. Looking forward to the next danis506 release. In the meantime
your workaround works quite well with my Revo64 Card (XFX's continued
production of the netcell-chip).

config.sys:
CALL=C:\OS2\CMD.EXE /Q /C C:\setlvm.cmd

setlvm.cmd:
diskinfo v
lvm /setnameartition,2,ECS_DATEN,ECS_DATEN
chkdsk e: /f /c

Seems that "diskinfo v" initializes the controller after first call
correctly.
--
Michael Holzapfel

On Mon, 30 Apr 2007 06:48:05 UTC in comp.os.os2.setup.misc, Michael Holzapfel
wrote:

> In the meantime
> your workaround works quite well with my Revo64 Card (XFX's continued
> production of the netcell-chip).

The dealer in the UK that sells these cards has end-of-lifed both 5 and 3 port
versions so I think XFX may just be using up stock they have on hand rather than
continuing to make the chips themselves. Anyone that wants one should probably
get one now - Scan in the UK only have the 3 port version left but they're only
£30! (plus shipping etc).

--
Trevor Hemsley, Brighton, UK
Trevor dot Hemsley at ntlworld dot com