[Novalug] Loaded question about SANs, but here goes

Brian Steisslinger brian.steisslinger@gmail.com
Tue Mar 20 12:41:47 EDT 2007


Do you get to determine raid set memebers across shelves and futher across
backend loops? How many backend loops are the filers?

I know a bunch of people who love the NetApp boxes I've just never had one
to play with... =-(

On 3/20/07, Ed T. Toton III <bones@necrobones.net> wrote:
>
> Thus spake Michael Stone:
>
> > If you're in a high availability market you're generally going to assume
> > that your disk will become unavailable regardless of how much you pay
> > for it. (E.g., you plan for an alternate site in case the first one
> > burns down.) Then you have some kind of higher-level logic to handle
> > whatever transition you need (like having service picked up by the
> > alternate site).
>
>
> That's definitely true. You have to assume disks and other components will
> fail or otherwise become unavailable, so I think a large part of the
> decision process should go into how much reliability and availability is
> going to be required.
>
> I know a lot of good discussoin has already gone by, and NetApp has been
> mentioned already, but I'll put a plug in for them. We use NetApp in our
> services. For us, HA was very important, as well as replication to a DR
> site (which can be handled seamlessly at the filesystem level by the
> netapps themselves).
>
> With a pair of NetApp heads, you can run them in an active-active
> redundant mode with redundant connections to the network switches as well
> as the disk shelves. In this case, we run a pair of disk volumes. The
> mission critical one is replicated to the other netapp, and the less
> critical is not. In our configuration, we can lose a disk, a switch, a
> netapp head, a NIC, a power supply, a FC module, even a disk shelf
> entirely, and the total downtime for automatic recovery is only about 15
> seconds. The only exception is if an entire shelf goes down for the
> non-critical volume, leaving that volume down until we can repair it. But
> that's a highly unlikely scenario, since there's not much to go wrong with
> a disk shelf. They have redundant power supplies and FC modules.
>
> These things just run, and can largely be treated as appliances. Set it
> and forget it. And the performance is also quite good.
>
> NetApp, in my experience, lives up to what they advertise. If that sort of
> high availability matters in your application, something similar might be
> appealing to you, if you can swallow the price tag. :)
>
>
>
> ------------------------------------------------------------------
> - Ed T. Toton III, RHCE --|-- www.necrobones.com -- ed.toton.org -
> ------------------------------------------------------------------
> Airplane Rule:
>         Complexity increases the possiblity of failure; a twin
>         engine airplane has twice as many engine problems as a
>         single-engine airplane.
>
>         (therefore the correct way to build a reliable system is
>         to put all your eggs in one really *good* basket)
>
> _______________________________________________
> Novalug mailing list
> Novalug@calypso.tux.org
> http://calypso.tux.org/cgi-bin/mailman/listinfo/novalug
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.firemountain.net/pipermail/novalug/attachments/20070320/ef69cc85/attachment.htm>


More information about the Novalug mailing list