linux-raid.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* Moving controllers
@ 2004-03-22 14:25 Robin Bowes
       [not found] ` <Pine.LNX.4.44.0403221748490.13036-100000@coffee.psychology.mcmaster.ca>
  0 siblings, 1 reply; 3+ messages in thread
From: Robin Bowes @ 2004-03-22 14:25 UTC (permalink / raw)
  To: linux-raid

Hi,

I have a question about moving drives between controllers with software RAID.

I'm planning on building a linux box with a large software RAID5 array (4
x 250/300GB disks) on which to store rips of my CD collection.

The server I have earmarked for this has built-in to it a "HighPoint
HPT370 PCI Dual Channel Ultra DMA/ATA 100 RAID Controller". I plan to
connect my four disks to the two ports of this controller, i.e. I will
have a master and slave drive on each channel.

First question: what is the kernel support like for this controller?

Secondly, I am aware that this is not the ideal configuration, but my
performance requirements are not massive as I won't be doing much more
than streaming MP3s to clients on my home network. However, if I were to
discover that performance was unacceptable I want to know if it would be
possible to install an additional controller in a PCI slot, and move the two
slave disks onto the new controller without re-building the array.

So, expressing my question explicitly:

Can I initially create a RAID5 array on this disk configuration:

HPT370--+--Channel A--+--master (Disk1)
        |             |
        |             +--slave (Disk2)
        |
        +--Channel B--+--master (Disk3)
                      |
                      +--slave (Disk4)

...then move the disks around like this:

HPT370--+--Channel A--+--master (Disk1)
        |
        +--Channel B--+--master (Disk3)

FooBar--+--Channel A--+--master (Disk2)
        |
        +--Channel B--+--master (Disk4)

...or this:

FooBar--+--Channel A--+--master (Disk1)
        |
        +--Channel B--+--master (Disk3)
        |
        +--Channel C--+--master (Disk2)
        |
        +--Channel D--+--master (Disk4)

...or even this:

FooBar 1--+--Channel A--+--master (Disk1)
          |
          +--Channel B--+--master (Disk3)

FooBar2--+--Channel A--+--master (Disk2)
         |
         +--Channel B--+--master (Disk4)


...without trashing the array?

Third question: If I were to decide on a replacement 4-port controller,
what models are well-supported?

Finally, I'd be interested to learn of any issues I might run into when
dealing with disks and filesystems of this size. For example, are there
any inherent limitations or absolute maximums I need to be aware of? What
about low-level issues like chunk size and block size, etc.? Any
recommendations as to the optimum settings, bearing in mind I will be
dealing with relatively large files?

Thanks for all comments.

R.
-- 
http://robinbowes.com

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: Moving controllers
       [not found]   ` <Pine.LNX.4.44.0403221748490.13036-100000@coffee.psychology.mcmaster.c a>
@ 2004-03-23 10:22     ` Robin Bowes
  2004-03-23 11:20       ` Brad Campbell
  0 siblings, 1 reply; 3+ messages in thread
From: Robin Bowes @ 2004-03-23 10:22 UTC (permalink / raw)
  To: Mark Hahn; +Cc: linux-raid

Hi Mark,

On Mon, March 22, 2004 22:53, Mark Hahn said:
>> connect my four disks to the two ports of this controller, i.e. I will
>> have a master and slave drive on each channel.
>
> this will cause anomalous performance in some circumstances,
> since master and slave are not concurrent.

Yeah, I read that in the list archives after I'd posted this message.

>> First question: what is the kernel support like for this controller?
>
> mediocre - the controller isn't generally as good as promise,
> nor is the support.

OK.

>
>> Secondly, I am aware that this is not the ideal configuration, but my
>> performance requirements are not massive as I won't be doing much more
>> than streaming MP3s to clients on my home network. However, if I were to
>
> which is very low data rate, what, under 200 KB/s?  since a floppy can
> do this, there is no performance issue with disks, no matter how badly
> configured.

:-)

I will doing other things with the box; for example, this box will also
host my mail, web server, databases, etc. None of this will require
spectacular performance, though.

>> possible to install an additional controller in a PCI slot, and move the
>> two
>> slave disks onto the new controller without re-building the array.
>
> this might work.  linux raid seems to want to have partitions marked 0xfd,
> and it writes some kind of self-identifying superblock onto each.
>
> why not just experiment and find out?  put two disks on a single channel,
> make the raid, power down, move the slave to another channel, and see if
> the raid assembles.

One word: time; I was hoping someone could tell me if it would work to
save me messing around. Of course, if I plan to do this I will perform
some sort of test first.

>
>> Third question: If I were to decide on a replacement 4-port controller,
>> what models are well-supported?
>
> promise.  3ware works well, but is silly expensive.

Yeah, I've seen that. What about the MegaRAID i4 ? That appears to do
hardware RAID5 which would save me having to bother with software RAID. Do
you know if it is true hardware RAID? Does it have good linux support?

>> dealing with disks and filesystems of this size. For example, are there
>> any inherent limitations or absolute maximums I need to be aware of?
>> What
>
> 2TB block-dev limit < 2.6.

What's the limit in 2.6? What about filesystem size?

>> about low-level issues like chunk size and block size, etc.? Any
>
> for mp3's, it cannot possibly matter.
>
>> recommendations as to the optimum settings, bearing in mind I will be
>> dealing with relatively large files?
>
> more than the 3-5 MB for a typical song?

I'll actually be using flac, and have a lot of classical music which has
movements up to 20 mins long  resulting in larger file sizes.

I read somewhere that XFS is robust and fast, but performance degrades
when the filesystem holds a lot of small files. When I say "relatively
large" I was meaning that I won't have a lot of small (eg. < 5k) files.

Thanks for the reply.

R.
-- 
http://robinbowes.com

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: Moving controllers
  2004-03-23 10:22     ` Robin Bowes
@ 2004-03-23 11:20       ` Brad Campbell
  0 siblings, 0 replies; 3+ messages in thread
From: Brad Campbell @ 2004-03-23 11:20 UTC (permalink / raw)
  To: Robin Bowes; +Cc: Mark Hahn, linux-raid

Robin Bowes wrote:

>>why not just experiment and find out?  put two disks on a single channel,
>>make the raid, power down, move the slave to another channel, and see if
>>the raid assembles.
> 
> 
> One word: time; I was hoping someone could tell me if it would work to
> save me messing around. Of course, if I plan to do this I will perform
> some sort of test first.
> 

The answer is yes it will work fine.
I had 5 200gb WD disks set as autodetect raid-5
My internal VIA IDE interfaces were hda/b/c/d
My external CMD680 IDE interfaces were hde/f/g/h

I had raid disks on c/d/e/f/g.
I created the raid and ran it for months, then I shuffled the disks and master/slave jumpers.
No problems.
I then purchased a couple of Highpoint 1540 SATA cards and a bundle of addonics SATA-> IDE adaptors.
Using the in-kernel hpt ata driver, no problem.
I then moved across to a pair of Promise SATA150-TX4 cards with the libata driver, no problem.

So as you can see, these disks have been "everywhere man" and I never had a problem with the kernel 
or appropriate tools picking them up in any order.

Brad

^ permalink raw reply	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2004-03-23 11:20 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2004-03-22 14:25 Moving controllers Robin Bowes
     [not found] ` <Pine.LNX.4.44.0403221748490.13036-100000@coffee.psychology.mcmaster.ca>
     [not found]   ` <Pine.LNX.4.44.0403221748490.13036-100000@coffee.psychology.mcmaster.c a>
2004-03-23 10:22     ` Robin Bowes
2004-03-23 11:20       ` Brad Campbell

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).