[Top] [All Lists]

Re: Software RAID, a bit OT

To: Simon Matter <simon.matter@xxxxxxxxxxxxxxxx>, Ben Gollmer <ben@xxxxxxxxxxxx>
Subject: Re: Software RAID, a bit OT
From: Seth Mos <knuffie@xxxxxxxxx>
Date: Fri, 19 Jul 2002 09:47:49 +0200
Cc: XFS <linux-xfs@xxxxxxxxxxx>
In-reply-to: <3D37B5F4.D799185C@xxxxxxxxxxxxxxxx>
References: <B95B1F8C.1E61%ben@xxxxxxxxxxxx> <3D36C9C3.BDFACE39@xxxxxxxxxxxxxxxx> <1027002434.5457.26.camel@xxxxxxxxxxxxxxxxxxxx> <200207181551.49007.ben@xxxxxxxxxxxx>
Sender: owner-linux-xfs@xxxxxxxxxxx
At 08:47 19-7-2002 +0200, Simon Matter wrote:
> I'm planning to use RAID 0+1 with XFS on kernel 2.4.18 (or 2.4.19 if it's out > by the time our disks come in). If there's any other good patches that you'd
> recommend for a server, let me know - I'm currently in the habit of running
> mostly vanilla kernels on my servers (except for XFS of course ;).

I believe there are some NFS patches floating around which some people recommend. It might be worth checking those out.

Now, another idea comes in because you want to be able to expand
existing arrays. I don't recommend LVM (LVM,EVMS,...) at the moment
because it makes things more complicated and it isn't worth it if you
only want to expand filesystems later by adding more disks. I recommend
using a linear 'RAID' on top of the RAID1 volumes. To expand later, you
may just create another RAID1 volume, append it to the linear array
while it is stopped and then grow the XFS filesystem on it. WARNING: I
have not tested linear arrays, if it works like the other RAID types, it
should be okay.

I like the theory but someone needs to prove it and let us know.

Normally you would need a abstraction layer like lvm for other raid levels to make this work.

The linear option does kill performance to that of single disk and you don't get the added write speed of the extra spindles. The read speed could be up to twice the speed of single disk.

It might just be your lucky day, if you only knew.

<Prev in Thread] Current Thread [Next in Thread>