I scanned through, but maybe I missed, bug was it ever determined what's
actually going onto this 12TB LUN? Aside from our longings for kilobytes vs
terabytes, and concerns of 75M files in a filesystem, I didn't see for sure.
An example:
I have a customer with 14TB on one sys. It's something like 64 LUNs, though
huge is huge.
Anyway, the 14TB was used for Oracle. Their backup strategy was to copy it to
an NFS mounted DataDomain box via 2 gigabit ports.
Last I checked, their backup times were on the order of 7 days, but only enough
disk space for 3 days of logs.
I think they were moving to a 10gbit DD box and were hoping that would solve
their issues.
Earlier recommendation was RMAN multiple channels into TSM, either direct fibre
or through virtual ethernet.
They were a recent conversion from HP and were pretty established on OmniBack.
TSM wasn't really an option (but neither was Omniback?)
Also, resource constraints were abound (Spec'd for 5 systems and a bunch of
4-port HBAs, deployed as 4-systems with a bunch of single and a couple of dual
port HBAs).
############
If journaling isn't an option, and it's multi-millions of files, then many
parallel backups. Either virtualmountpoints or separate filesystems. A
separate scheduler for each filesystem, resourceutil 10 on each one, and decent
CPU and I/O bandwidth to handle the scans.
I found that on the above customer's DMX, we could get about 96kIOPS arraywide.
From the DB system, we could get 64kIOPS, but since 40-50 other LPARs were
deployed onto it, plus vmware, etc, the usable IOPS was around 28k. Stat for a
file is an I/O, so scanning a large number of files needs a large IOPS support.
If it's a *NIX, it might be worth using some sort of cluster filesystem. Then
you could put the metadata on decent SSDs for higher IOPS on scans...
Lots of options.
With friendly regards,
Josh-Daniel S. Davis
----- Original Message ----
From: "Gill, Geoffrey L." <GEOFFREY.L.GILL AT SAIC DOT COM>
To: ADSM-L AT VM.MARIST DOT EDU
Sent: Wed, May 26, 2010 6:03:38 PM
Subject: [ADSM-L] why create a 12TB LUN
I'm guessing many of you will find this quite odd, I know I did, but I
had someone come to me and say they were going to ask for a 12TB LUN and
wanted to back it up. Without even mentioning the product they want to
use, obviously not TSM though, and I'm not even sure it would make
difference, how would you manage to get a 12TB LUN backed up daily. I
would expect it to be at least 75% full if not more, and even without
knowing what percentage of data changes on it, it would seem to me the
request seems strange. They're thinking of getting a VTL and backing up
through fiber direct, not across the network, but no idea which one or
what sort of throughput to expect.
Have any of you been approached with this sort of request and if so what
was your response? I'm sort of dumbfounded at this point since I've not
heard or seen this anywhere.
Thanks,
Geoff Gill
TSM/PeopleSoft Administrator
SAIC M/S-B1P
4224 Campus Pt. Ct.
San Diego, CA 92121
(858)826-4062 (office)
(858)412-9883 (blackberry)
|
