05-30-04 04:10 PM
Thanks Nik.
Ok here's where we're at.
We've got 3 dell poweredge servers that are running an ASP.NET app. Thats
connecting to a SQL Server 2000 standalone. The app we're running involves
uploading significant numbers of bytes from teh clients to be stored as
files somewhere.
We need high availability and centralised storage. What I am trying to do
is have failover clustering working on both the database server and the
webservers.
What I was hoping was that the database servers could be in a failover pair
using a shared hard-disk enclosure and that the 3 web-servers could also be
attached to that one hard-disk enclosure.
The space requirements are geared toward the web-servers. We'll be
accepting about 200 gb of files from users using the applciation - whereas
the SQL database will only be a few gigabytes. So the space requirements of
teh file servers are much higher.
Cheers,
Ken
"Nik Simpson" <n_simpson@bellsouth.net> wrote in message
news:n5afc.13887$Yw5.12017@bignews4.bellsouth.net...
> Ken Shaw wrote:
>
> Well, yes and no, the main reason that SCSI never really took off as the
> basis for shared storage is that it doesn't work particularly well. You
can
> build a small shared storage cluster around SCSI, but you may well end
with
> more trouble than you expect with performance in particular.
>
>
> Clustering in the Windows world is "shared nothing" i.e. if you have two
> machines in a cluster, machine 1 exclusively owns its filesystems and
> machine 2 exclusively owns its filesystems. If for example machine 1
fails,
> machine 2 takes over the filesystems (and associated applications.)
>
> There is no general mechanism in Windows clustering for shared access to a
> single filesystem at the SCSI/FC level that requires additional software
> such as the examples mentioned by Rob.
>
> I don't beleive that MS supports shared SCSI storage in Windows clustering
> anymore because of the problems everybody had trying to get it to work in
> the late 90s (I know, I was one of the poor slobs trying :-) These days,
MS
> clustering (and anybody else's clustering for that matter) pretty much
> assumes Fibre attached storage.
>
set[vbcol=seagreen]
>
> Define "direct attached" there is nothing in the Windows clustering model
> that requires direct attached as opposed to SAN attached because
> functionally (i.e. to the OS) there is no difference between having an
> exclusive point-to-point connection from the host to a Fibre array and
doing
> the same thing via an FC switch.
>
>
> If you mean can it handle the locking issues associated with a shared
> filesystem, the answer is no, it doesnt even try.
>
> Perhaps you could explain the "high-level" problem that is leading you
> towards a shared SCSI storage solution, there may be tother ways of
> addressing the problem.
>
>
> --
> Nik Simpson
>
>
[ Post a follow-up to this message ]
| 
