Very Computer

by **Kueh, Anthon** » Sat, 06 Apr 1996 04:00:00

On a lot of higher end performance servers, they are able to stripe the
hard drives to create a file system which spans over multiple physical
hard drives. (ie. /dev/vda1 spans over cyl 0-500 on HD0 and HD1) This
configuration provides a dramatic increase in system performance due to
the fact that there are physically two heads handling the data retrieval
for one file system. I was wondering if there's such a thing for Linux,
and/or if not, is anyone developing this..

-Tony

-------------------------------
Anthony Kueh
Department of Computer Sciences

by **Christoph Lamet** » Sat, 06 Apr 1996 04:00:00

: On a lot of higher end performance servers, they are able to stripe the
: hard drives to create a file system which spans over multiple physical
: hard drives. (ie. /dev/vda1 spans over cyl 0-500 on HD0 and HD1) This
: configuration provides a dramatic increase in system performance due to
: the fact that there are physically two heads handling the data retrieval
: for one file system. I was wondering if there's such a thing for Linux,
: and/or if not, is anyone developing this..

Its part of the 1.3.X kernels. Never ask if Linux can - always look at
the recent releases or ask how to do it.

by **Stefan Bernhar** » Sun, 07 Apr 1996 04:00:00

: Its part of the 1.3.X kernels. Never ask if Linux can - always look at
: the recent releases or ask how to do it.

Ok. How do you do it? ;)

--
--"Color's for end-users."
--Stefan Bernhardt, Vice-President ACM

by **Rhys Thury** » Tue, 09 Apr 1996 04:00:00

: : Its part of the 1.3.X kernels. Never ask if Linux can - always look at
: : the recent releases or ask how to do it.

: Ok. How do you do it? ;)

Compile in the md (meta-disk) driver in a newer 1.3.x kernel (I'm using
1.3.83 successfully, but I don't use md). Then get the md034 package from:

ftp://sweet-smoke.ufr-info-p7.ibp.fr/public/Linux/md034.tar.gz

This information was found in the kernel source tree in
/usr/src/linux/drivers/block/README.md.

--------
If you can lead it to water and force it to drink, it isn't a horse.

Got a Linux problem? Or can you help others solve them? Visit the Linux
Common Problems page at http://vortex.cc.missouri.edu/~rhys/linux.html

by **Jan Rychte** » Fri, 12 Apr 1996 04:00:00

Anthony> On a lot of higher end performance servers, they are able
Anthony> to stripe the hard drives to create a file system which
Anthony> spans over multiple physical hard drives. (ie. /dev/vda1
Anthony> spans over cyl 0-500 on HD0 and HD1) This configuration
Anthony> provides a dramatic increase in system performance due to
Anthony> the fact that there are physically two heads handling the
Anthony> data retrieval for one file system. I was wondering if
Anthony> there's such a thing for Linux, and/or if not, is anyone
Anthony> developing this..

Yes, there is, it's called md. But, has anyone actually gotten it to
work reliably ? For me, it has locked my machine dead in a couple of
minutes, ever since md-0.33 through its integration into the
mainstream kernel and up to 1.3.84. I have two WD IDE drives that I
otherwise use heavily all the time, with no problems. It's only with
the md driver that problems occur.

jwr
___________
Jan Rychter

by **Mikael Abrahamss** » Sat, 13 Apr 1996 04:00:00

Quote:
> Yes, there is, it's called md. But, has anyone actually gotten it to
>work reliably ? For me, it has locked my machine dead in a couple of
>minutes, ever since md-0.33 through its integration into the
>mainstream kernel and up to 1.3.84. I have two WD IDE drives that I
>otherwise use heavily all the time, with no problems. It's only with
>the md driver that problems occur.

I have used disk concatenation (LINEAR) with 1.3.71 to make a 2.2 gig
partition out of two Quantum Fireball EIDE drives (1080 and 1280). This
has worked flawlessly ever since, and it seems very stable.

I wouldnt use RAID0 with EIDE though, I consider the DMA-use too high for
that. Using it to just linear two drives together I see no problem though.
--
-----
Mikael Abrahamsson

Student vid Institutionen f|r Informatik vid Lunds Universitet, Sweden.

by **Kueh, Anthon** » Sat, 13 Apr 1996 04:00:00

> Anthony> On a lot of higher end performance servers, they are able
> Anthony> to stripe the hard drives to create a file system which
> Anthony> spans over multiple physical hard drives. (ie. /dev/vda1
> Anthony> spans over cyl 0-500 on HD0 and HD1) This configuration
> Anthony> provides a dramatic increase in system performance due to
> Anthony> the fact that there are physically two heads handling the
> Anthony> data retrieval for one file system. I was wondering if
> Anthony> there's such a thing for Linux, and/or if not, is anyone
> Anthony> developing this..

> Yes, there is, it's called md. But, has anyone actually gotten it to
> work reliably ? For me, it has locked my machine dead in a couple of
> minutes, ever since md-0.33 through its integration into the
> mainstream kernel and up to 1.3.84. I have two WD IDE drives that I
> otherwise use heavily all the time, with no problems. It's only with
> the md driver that problems occur.

I've had the same luck, with Kernel 1.3.6x .... It locked my machine
after I start to do some heavy disk accessing. From what I've read, this
is still an alpha package. Anyone actually had real success with this?

-Tony

-------------------------------
Anthony Kueh
Department of Computer Sciences

by **Kueh, Anthon** » Sat, 13 Apr 1996 04:00:00

Disk Concatenation defeats the purpose of the true performance benefits
md can give you. There are no performance gains between md in
concatenation mode and regular good old e2fs.

-Tony

> > Yes, there is, it's called md. But, has anyone actually gotten it to
> >work reliably ? For me, it has locked my machine dead in a couple of
> >minutes, ever since md-0.33 through its integration into the
> >mainstream kernel and up to 1.3.84. I have two WD IDE drives that I
> >otherwise use heavily all the time, with no problems. It's only with
> >the md driver that problems occur.

> I have used disk concatenation (LINEAR) with 1.3.71 to make a 2.2 gig
> partition out of two Quantum Fireball EIDE drives (1080 and 1280). This
> has worked flawlessly ever since, and it seems very stable.

> I wouldnt use RAID0 with EIDE though, I consider the DMA-use too high for
> that. Using it to just linear two drives together I see no problem though.
> --
> -----
> Mikael Abrahamsson

> Student vid Institutionen f|r Informatik vid Lunds Universitet, Sweden.

-------------------------------
Anthony Kueh
Department of Computer Sciences

by **Stormy Henders** » Sat, 13 Apr 1996 04:00:00

-----BEGIN PGP SIGNED MESSAGE-----

Disk Concatenation defeats the purpose of the true performance
benefits md can give you. There are no performance gains between md
in concatenation mode and regular good old e2fs.

If you can't have the whole cake, you'll refuse to even eat a slice? DC is
a very nice feature, if I had more than one drive I would love it, but if
striping isn't yet ready for primetime, I certainly wouldn't refuse to use
any of the other nice benefits that MD has to offer.

Be happy...

http://www.gtlug.org/~stormy/signature

-----BEGIN PGP SIGNATURE-----
Version: 2.6.3
Charset: ascii

iQCVAgUBMW6n07urJgxna7opAQFEBQP/ZipE/d4yiqTQP2l4MznNmVKB8TxNByRn
3yptV529BHL6TY6kNXPNwgzTygueOxflqeiwzv9ZMGaoa+zebb5Vjyd4fWgRPjLE
YB7muwjBDt7z0/7VK36d1jOAeYs2oo0Mriz0WQKfJw1J/4pYwRdIR3/EmEpkuMUY
wVsieMk3MmQ=
=0Us9
-----END PGP SIGNATURE-----

by **Mikael Abrahamss** » Sat, 13 Apr 1996 04:00:00

>Disk Concatenation defeats the purpose of the true performance benefits
>md can give you. There are no performance gains between md in
>concatenation mode and regular good old e2fs.

Performance was never the issue for me, large partitions were. I'd rather
have a 2.2 gig partition than two of little over 1 gig each.

--
-----
Mikael Abrahamsson

Student vid Institutionen f|r Informatik vid Lunds Universitet, Sweden.

by **Rhys Thury** » Wed, 17 Apr 1996 04:00:00

: Disk Concatenation defeats the purpose of the true performance benefits
: md can give you. There are no performance gains between md in
: concatenation mode and regular good old e2fs.

Except that you then have a single filesystem instead of two. Suppose I
have a large directory structure that I want to move to a separate
device, but it's very smoothly distributed. I can't just choose one
section of it for one device and put the rest on another. I need to put
the whole thing on one big device (or suppose I have some >300 MB files,
but little hard drives). There's no *performance* gain, but at least it
works.

If you can lead it to water and force it to drink, it isn't a horse.

Got a Linux problem? Or can you help others solve them? Visit the Linux
Common Problems page at http://vortex.cc.missouri.edu/~rhys/linux.html

by **Adam Megac** » Fri, 19 Apr 1996 04:00:00

> : Disk Concatenation defeats the purpose of the true performance benefits
> : md can give you. There are no performance gains between md in
> : concatenation mode and regular good old e2fs.

> Except that you then have a single filesystem instead of two. Suppose I
> have a large directory structure that I want to move to a separate
> device, but it's very smoothly distributed. I can't just choose one
> section of it for one device and put the rest on another. I need to put
> the whole thing on one big device (or suppose I have some >300 MB files,
> but little hard drives). There's no *performance* gain, but at least it
> works.

I hear that md is unstable. Is disk concatenation any more stable? If so, where can I get
the disk-concatenation driver?

by **Mikael Abrahamss** » Mon, 22 Apr 1996 04:00:00

>> Except that you then have a single filesystem instead of two. Suppose I
>> have a large directory structure that I want to move to a separate
>> device, but it's very smoothly distributed. I can't just choose one
>> section of it for one device and put the rest on another. I need to put
>> the whole thing on one big device (or suppose I have some >300 MB files,
>> but little hard drives). There's no *performance* gain, but at least it
>> works.

>I hear that md is unstable. Is disk concatenation any more stable? If
> so, where can I get the disk-concatenation driver?

disk concatenation is the LINEAR option in md. I've used it for over a
month now with 1.3.71, and it has been very stable. I striped two large
EIDE drives together two form one big partition.

I kinda got the notion that EIDE isnt good for RAID0, so I wouldnt do that
myself.
--
-----
Mikael Abrahamsson

Student vid Institutionen f|r Informatik vid Lunds Universitet, Sweden.

by **Rhys Thury** » Tue, 23 Apr 1996 04:00:00

: > Except that you then have a single filesystem instead of two. Suppose I
: > have a large directory structure that I want to move to a separate
: > device, but it's very smoothly distributed. I can't just choose one
: > section of it for one device and put the rest on another. I need to put
: > the whole thing on one big device (or suppose I have some >300 MB files,
: > but little hard drives). There's no *performance* gain, but at least it
: > works.

: I hear that md is unstable. Is disk concatenation any more stable? If so, where can I get
: the disk-concatenation driver?

md *is* the disk-concatenation driver. It can work in three modes:
concatenate (linear), RAID 0 (striping), and RAID 1 (mirroring?).
Anyway, I think using in in linear mode is supposed to be more stable
than using it in RAID 0 mode.

--------
If you can lead it to water and force it to drink, it isn't a horse.

Got a Linux problem? Or can you help others solve them? Visit the Linux
Common Problems page at http://vortex.cc.missouri.edu/~rhys/linux.html

Very Computer

Hard Disk striping

Hard Disk striping

Hard Disk striping

Hard Disk striping

Hard Disk striping

Hard Disk striping

Hard Disk striping

Hard Disk striping

Hard Disk striping

Hard Disk striping

Hard Disk striping

Hard Disk striping

Hard Disk striping

Hard Disk striping

Hard Disk striping