|
I walked this path a few months ago, though with different design constraints.
I ran into a lot of the same stuff you did: everybody is SAN-crazy. I didn't
like the thought of having to rely on NFS (or worse) for my production data
access, so we tried to find a practical SCSI-attached solution.
A few companies do make SCSI/RAID cabinets which will dual-attach. Mylex
reportedly makes one, although I had zero luck finding a reseller who even
knew WTF I was talking about, let alone being able to send me a quote. Compaq
makes one, it looked good but was outside our (small) budget. And of course
Sun makes stuff like this. We also found a company called "Dynamic Network
Factory" which, aside from having one of the most buzzword-compatible names
I've seen, looked like they made something which would do this for cheaper
than anybody else.
http://www.dynamicnetworkfactory.com/RAID/PowerStorTower.htm
You didn't say this specifically, but if you want multiple attached hosts
to access the same disk volumes read-write, you may be out of luck for now.
We wanted that, and ended up having to drop back to good ol' NFS to get it
done. The dual-attach cabinets don't have any internal logic for brokering
write access, they assume that you'll either a) group the disks for access by
one host or the other, b) that only one host will have write access across the
volumes while the other is read-only, or c) that you'll have some software
process on the attached hosts which governs shared write access while avoiding
data corruption. Well, there are projects in the works to do this, but as
far as I could tell none of them are ready yet. Sun resells one but their rep
was unwilling to recommend it to me for this purpose, and anyhow it won't run
on Linux. The Linux LVM project has this sort of functionality on their "to
do" list. Other manufacturers may have something (SGI sounds like one likely
source), but that doesn't do us penguinistas any good until they finish porting
their own logical volume software to Linux, and maybe not even then.
HTH,
-m
On Mon, Oct 09, 2000 at 06:25:46PM -0700, Dan Browning wrote:
> Sorry for the off topic post:
>
> I'm building a Linux cluster here soon. I'm looking at storage options for
> my database servers and web servers. I *could* put 10-disk arrays in each
> database server, but I'm looking for an option that would allow me to
> consolidate all those disks and storage and have multiple Ultra-160
> interfaces to those disks.
>
> The benefit would be that all I have to buy is one ultra-160 controller for
> each server, and connect it to the "main storage", and the main storage
> would take care of doing the RAID-5 on 5, 10, or 20 discs. To the server it
> would just look like one big fast SCSI disk.
>
> I've read into EMC, brocade, gadzoox, hitachi, etc. etc. But all I find is
> SAN stuff (fiber channel interfaces, etc.). I would prefer Ultra-160 over
> fiber channel because I don't need the distance of FC and I prefer the
> stability of Ultra-160 SCSI card drivers and simplicity.
>
> Does anyone have any pointers to information or resources that I could
> research?
>
> Dan Browning
> Network & Database Administrator
> Cyclone Computer Systems
>
>
>
> _______________________________________________
> Redhat-list mailing list
> Redhat-list@xxxxxxxxxx
> https://listman.redhat.com/mailman/listinfo/redhat-list
--
Michael Jinks, IB
Systems Administrator, CCCP
finger mjinks@xxxxxxxxxxxxxxxxxxxxxxx for public key
Vote Duke! http://www.entertaindom.com/pages/duke2000/home.jsp
|
