Fibre Channel disks to two Systems?

Danny Braniss danny at cs.huji.ac.il
Sat Sep 10 01:39:13 PDT 2005


hi danny
	you are asking too many questions :-), but w/r to netapp:
same computer, 1gbE, NFS is about 50% slower than FC.
btw, iSCSI (still beta) is only slightly faster than NFS
(note NFS is UDP, iSCSI is TCP).

as to reliability, the netapp is worth avery penny (actualy K$ :-), had only 
one major breakdown in over 10 years.

for backup, concider (depends on the size of your database), copying the WAL
to the backup host/disc, and running it at some interval to update the backup.

and another thing, from experiance, disks break more often than cpus, so 
offloading
the db might give you only 'some' backup.

	danny


> [NOTE: If posting followup, please mind the
>        cross-post to -questions and -scsi.]
> 
> Hello,
> 
> We host our PostgreSQL database on FreeBSD.  Until now, we have just
> built the beefiest DB server we can spec, and then dump the data every
> thirty minutes to a backup DB server, so if the primary DB server fails,
> we load the database on the backup and fail over to the backup server.
> 
> But I'd rather offload the disk to an external storage device, then I
> can have two identical DB servers, and if one fails, I swap the disks
> over to the other DB server, mount the filesystem, possibly run data
> consistency checks, and proceed from there.
> 
> >From my research, I am thus far most impressed with the SANbloc 2Gb,
> which holds fourteen FC drives in a 3U rackmount.  It can be had with
> redundant RAID controllers, or as a JBOD.  There are similar products
> from other vendors as well.
> 
> I could concievably do the RAID in software by running a gstripe across a
> set of gmirrors.
> 
> As I understand it, I can have an FC loop with one or more drives,
> connected to two servers, and either server can talk to one or the other
> drives exclusively.  My QUESTION is: how is the arbitration done in
> FreeBSD?  You run camcontrol on either server and activate / deactivate
> drives in the loop?
> 
> What happens if say, the primary server locks up in some weird manner?
> Can it block the backup server from talking to the drives?  (We can
> always have a NOC tech turn off a badly failed primary database, and
> power-cycle the disk array, if needed ...)
> 
> A really far-out idea I had was that with fourteen drive bays I could
> have two hot spares, and then set up a stripe across four mirrored pairs
> (4x2 = 8-disk RAID10) and then with the remaining four drives assign
> each to be a third component of the gmirrored pairs, let the gmirrors
> sync up, then detach those drives from the gmirrors, mount them on the
> backup database, gstripe those containers together, and have a
> point-in-time "snapshot" of the drive array that could be mounted on the
> backup server, from which I could run database dumps, or conduct
> failover tests, etc.  (I could kick this around -geom. :)
> 
> Uhmmm, has anyone done similar?  Suggestions?  Feedback?  Advice?
> 
> Or, should I try to get a NetApp, or similar device, even though FreeBSD
> does not support iSCSI, because NFS performance over GigE may still beat
> FC?
> 
> Also, does anyone have a FreeBSD-friendly storage systems integrator or
> other vendor they can reccomend, particularly one near the San Francisco
> area?  I keep contacting various vendors who then fail to get back to
> me. :(
> 
> Thanks for all feedback and suggestions!
> 
> Sincerely,
> -danny
> 
> -- 
> http://dannyman.toldme.com/
> _______________________________________________
> freebsd-scsi at freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-scsi
> To unsubscribe, send any mail to "freebsd-scsi-unsubscribe at freebsd.org"
> 




More information about the freebsd-questions mailing list