I agree with your thoughts on performance but not the use of large raid groups. It would eat into the data spindle count but I think 10 disk RGs would increase the warm fuzzy feeling. I have pushed single parity RGs well past the safe point without problems but the guy is antsy about failures. I have never lost any data in the years I have been taking care of the filers.
The plan right now is to use what comes with Solaris for filesystem management. VxVM was thought of for a while.
Mike
-----Original Message----- From: slinkymax0r [mailto:slinkywizard@integra.net] Sent: Sunday, February 18, 2007 4:06 PM To: tmac Cc: Miller, Michael CTR USTRANSCOM J2; toasters@mathworks.com Subject: Re: Large Oracle database question
hmmm... with an adequately sized head, his work load shouldn't be a problem at all.
I would actually anticipate better performance/ recoverability with large aggregates of fast spindles, and RAID-DP means very large raid groups :)
Would you be using NFS or block? If block, what file system would you be using over the RAID array?
Regards, Max
Has anyone explained or showed this person a write-up on WAFL and NetApp's implementation of RAID-4 and now RAID-DP? ...how NetApp's version does not have a "hot" parity disk as all disks are excersized about the same? ...how RAID-DP is superior to RAID 1+0 and in an edge case is even better (if you loose the same disk in both plexs of a RAID 1, you have total data loss...you can loose any two disks with RAID-DP and loose nothing) ...how RAID-DP has very little overhead compared to NetApp's RAID-4? ...how NetApp's RAID-4 is faster than most hardware RAID arrays?
I could go on, and on, an on....(ask others that know me...plus 10 years at netapp helped)
-- --tmac
RedHat Certified Engineer
On 2/16/07, Miller, Michael CTR USTRANSCOM J2 michael.miller.ctr@ustranscom.mil wrote:
(This might show up twice) We are getting ready to stand up a 30+TB GIS app and I was wondering
if
anyone on the list has done something like this on a filer.
The guy who will be running this doesn't think a filer would give him the performance required (not specified) and is also concerned that the parity drives will get beat to hard.
He has been told by others that the only way to make this work as
needed
is to use RAID 1+0. This started out being required for all of the data
but
has been relaxed to just the indexes and current point of interest data.
Any thoughts?
Mike Miller General Dynamics Information Technology Michael.Miller.ctr@ustranscom.mil Michael.Miller@gdit.com Phone: 618-229-1185