Skip site navigation (1)Skip section navigation (2)
Date:      Thu, 5 Nov 1998 11:40:31 +1030
From:      Greg Lehey <grog@lemis.com>
To:        Steve Ames <steve@ns1.cioe.com>, steve@inc.net
Cc:        freebsd-isp@FreeBSD.ORG
Subject:   Re: Vinum on root partition (was: disk mirroring?)
Message-ID:  <19981105114031.S784@freebie.lemis.com>
In-Reply-To: <199811041513.KAA06958@ns1.cioe.com>; from Steve Ames on Wed, Nov 04, 1998 at 10:13:07AM -0500
References:  <19981104150946.L784@freebie.lemis.com> <199811041513.KAA06958@ns1.cioe.com>

next in thread | previous in thread | raw e-mail | index | archive | help
On Wednesday,  4 November 1998 at 10:13:07 -0500, Steve Ames wrote:
>> Ah, you're thinking in terms of ccd, where everything crashes if one
>> drive goes down (thus lowering uptime instead of increasing it).  If
>> you lose one plex with Vinum, the others are unaffected.  Here's a
>> case from my test machine:
>>
>> vinum -> l -r obj
>> V obj                   State: up       Plexes:       2 Size:        400 MB
>> P obj.p0              C State: up       Subdisks:     2 Size:        400 MB
>> P obj.p1              S State: down     Subdisks:     4 Size:        400 MB
>> S obj.p0.s0             State: up       PO:        0  B Size:        200 MB
>> S obj.p0.s1             State: up       PO:      200 MB Size:        200 MB
>> S obj.p1.s0             State: up       PO:        0  B Size:        100 MB
>> S obj.p1.s1             State: up       PO:      100 MB Size:        100 MB
>> S obj.p1.s2             State: up       PO:      200 MB Size:        100 MB
>> S obj.p1.s3             State: up       PO:      300 MB Size:        100 MB
>
> I have things setup to stripe and thats what confused me. In my mind if
> I take 2 200M partitions and stripe across them I should end up with
> 1 200M plex (the other 200M being completely a mirror state). I can't
> have 400M because then if 1 drive goes bad I'm out half my data
> right?

No, you're misunderstanding.  There is no data replication within a
plex.  Take 2 200 MB *subdisks* (Vinum doesn't talk about partitions)
and stripe them together and you get a 400 MB plex.  Put 2 400 MB
plexes in a volume and you have a 400 MB volume with mirroring.

> In your above example you have a striped plex with 4 subdisks. Each
> subdisk is 100M. The total capacity of the plex is 400M. If I write
> 400M of data to that and lose a subdisk don't I lose data?

You do on that plex (obj.p1 in this example)

> How is a "mirroring" setup configured?

That's what obj.p0 is for.  In this case, for example, the entire plex
obj.p1 is down, but the volume is still accessible.

>>> Then there the issue of no 'root' partition support.
>>
>> You said.
>>
>> Would you be interested if I implemented partition encapsulation?
>> Basically, that would be a subdisk which is not part of a Vinum drive,
>> so you could point to an existing disk partition.  This would enable
>> you to mirror in the way you planned, and would also make it easier to
>> support root partitions.
>
> Ah. Sweetness. If I understand you correctly here that would be most
> excellent. Obviously you'd lose the striping efficiencies here since
> one of the subdisks ins't part of the vinum drive, but that would let
> you mirror things in a very simple manner and then support dual
> writes.

Yes, there would be restrictions on the encapsulated partition.  It
would have to be concatenated, and it doesn't seem to make any sense
to have more than one subdisk in that particular plex.  The other
plexes in the volume could have any organization, of course.

> You could probably also still read from the most unused drive while
> writing to both, this would give _some_ efficiencies.

Sure, that's standard on Vinum.  It reads from only one plex, by
default in a round-robin fashion.  By contrast, ccd always reads from
the first of the pair.

Greg
--
See complete headers for address, home page and phone numbers
finger grog@lemis.com for PGP public key

To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-isp" in the body of the message



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?19981105114031.S784>