Can your SSA adapters do RAID10 ? If yes, you could think about it.
RAID10 should give you reasonably good performance (RAID0) and proteccion (RAID1)
[email]j_ammann[/email] (Joerg Ammann) wrote in message news:<a5fbda48.0307082225.7ba8187bposting.google. com>...> [email]sasa1973pk[/email] (sasa queer) wrote in message news:<702930ee.0307080034.4e3d8d3posting.google.c om>...>> > Dears,
> > I want some quick help from you all guys as customer now want me to
> > respond as soon as possible......and i think preobably no one in RISC
> > world has think forever for redundancy over ssa boxes...that's why i
> > did not get any response.
> > SASA
> > [email]sasa1973pk[/email] (sasa queer) wrote in message news:<702930ee.0307030358.2da58c86posting.google. com>...> > > Dears,
> > > I have a customer who is so much curious about redundancy matter that
> > > he thinks
> > > that a shared SSA between two RISC nodes is a single point of failure(
> > > inspite of having two hot spare disks).
> > > Now he wants to place another SSA box with same number of disks in the
> > > same ssa loop with existing SSA box.However he is afraid of
> > > performance issues!!!!!
> > > what i suggest that implement two raid5 on two controllers and then do
> > > mirroring on O/S level.
> > > Can anybody help me in predicting performance issues if there!!!!
> > > if yes then what will be other alternatives ???????please guide !!!
> > >
> > > SASA
> our setup is:
> - 2 pserver, each with 2 ssa-adapters
> - 2 ssa-box
> the loops are:
> server1/ssa-adapter1 --- ssa-box1 --- server2/ssa-adapter1 ---
> back-to server1
> server1/ssa-adapter2 --- ssa-box2 --- server2/ssa-adapter2 ---
> back-to server1
> the logical-volumes are mirrorred from ssa-box1 to box2
> this gives us redundancy for:
> - failure of ssa-disk or whole ssa-box
> - failure of ssa-adapter
> - and (with some manual work, as this is only a poor mans HACAMP) for
> failure of server
> theres some more work for the os io-subsytem, because of the
> write-doubling, but we never encountered a performance problem.