Greetings, ADSMites. (Hmm. Need a better collective name)
At the North East Regional Data Ceter here at the University of Florida, we're
about to implement disk pools in a serious way for the first time. I'm
interested in your comments or suggestion w.r.t. our plans.
our ADSM machine is a 4-processor, "wide silver" SP node, with 2G of ram. (A
UNIX box) We currently store our database and log on four SSA disks, with
dsmfmt'ed files mirrored by ADSM itself. I considered AIX-level mirroring, but
chose otherwise after seeing the discussions on this list.
We've got two major serial media pools: a 3995 with about 160 2.4G platters,
And a 3494 with ~150 volumes which we share with a MVS system. We're going to
be adding 8 more SSA disks tasked as disk pools.
Now, here's the complicated part.
We serve several groups within the Data Center, and are extending the use of
this server to an increasing number of units elsewhere in the University. As
this happens, we have created a burgeoning set of management classes and
administrators. Easy enough. But they have to have separate storage pools, so
we can give them authority to mess with their storage. So I need a burgeoning
set of disk pools, in a variety of sizes: our 'major server' node group needs
a bigger disk pool than most of the other units we serve.
With that as context:
I intend to take the 8 new ~9G SSA disks, toss them all into a volume group,
and then on each disk make the following raw logical volumes:
a: 4G volume
b: 2G volume
c: 1G volume
d: 1G-plus-whatever-slop volume.
I know, that doesn't add up to 9G, I may be able to get more real estate when
I create the new volume group; my test VG I used 256M physical partitions,
with the implicit waste of that tile size. If the 1G+slop is more like 2G,
then that's OK with me.
Each of these logical volumes is to be restricted to a single disk, so I can
be sure that if I assign three 'd' volumes to a stgpool, they'll be separate
spindles.
Once that is done, I intend to parcel the volumes out amongst my storage
pools, making sure that no pool has multiple volumes from the same spindle,
unless I decide (for some reason) that I need more than a full "stripe" across
my disk farm.
I'll try to avoid backup window overlap, but I may not be able to. In any
case, even if I'm backing up the servers to all of the 'a' stripe and the
department of left-handed tiddlywinks is backing up to some of the 'c' stripe,
it's unlikely that the disk write speed will be my throughput bottleneck; I'll
be able to support them both.
So here's an exemplar configuration for AIX disk pools. Am I sounding
reasonable? What can I do better?
Allen S. Rout
'ADSMphasis'? ADSMpire? ADSMpowerers? Hmmm.
PS: The U-nicks post was a troll.
|
