xfs
[Top] [All Lists]

Re: LVM on Linux

To: Florin Andrei <florin@xxxxxxx>
Subject: Re: LVM on Linux
From: Ragnar Kjørstad <xfs@xxxxxxxxxxxxxxxxxxx>
Date: Tue, 17 Jul 2001 00:03:59 +0200
Cc: linux-xfs@xxxxxxxxxxx
In-reply-to: <995318493.12677.14.camel@stantz.corp.sgi.com>; from Florin Andrei on Mon, Jul 16, 2001 at 02:21:33PM -0700
References: <85063BBE668FD411944400D0B744267A643419@AUSMAIL> <20010716225418.G14564@vestdata.no> <995318493.12677.14.camel@stantz.corp.sgi.com>
Sender: owner-linux-xfs@xxxxxxxxxxx
On Mon, Jul 16, 2001 at 02:21:33PM -0700, Florin Andrei wrote:
> On 16 Jul 2001 22:54:18 +0200, Ragnar Kjørstad wrote:
> > 
> > Even for highend RAID's there are lots of reasons to use LVM/EVMS:
> > * load balancing across multiple scsi-channels (not implemented yet)
> > * Dynamic partitioning
> > * Volumes spanning multiple physical devices
> > * snapshotting.
> > 
> > Basicly handling this is software adds a lot of flexibility.
> 
> But how about performance?
> Aren't these things supposed to be handled by the hardware?
> It's the same thing as for soft vs hard RAID. Hard should be faster
> (but, yeah, soft is easier to play with).

There is a performance penalty, but if you're using LVM just for
partitioning it's very small. (I believe ca 4% if you're using large
extents). 

I'd assume the penalty for doing striping is approxemately the same, but
haven't tested it. Basicly striping is a technology that give you better
benchmarks results but often doesn't improve real-life performance.
(because benchmarks are often a single process writing to a single file
while real-life is multiple users writing to multiple files - so your
load is distributed over the different devices anyway)

For mirroring and even more so for raid5, it's a different story. If you
want high performance, you don't want to do mirroring/raid5 in software.


-- 
Ragnar Kjorstad
Big Storage


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