I've embraced centralized storage in a big way with the hopes that it would 
facilitate the backup process (e.g. if you have all the data and boot images on 
one system you don't even worry about doing network backups).  In the last six 
months I've deployed 3 Solaris/ZFS storage servers to handle storage needs for 
my home office and 2 remote colos.  

I've been using Amanda for about a year now but have run into a number of 
problems when the volume of data starts scaling into the Terabytes.  I think 
Amanda is capable of dealing with much more, but I'm finding that the Best 
Practices for doing this aren't very well documented.  So I'm going to outline 
what I've been doing and what challenges I've been encountering in hopes that 
we all might be able to collectively improve the documentation for these kind 
of scenarios...which if increasingly important as more and more folks take the 
leap as I have to centralized storage architectures.


Here's what my infra looks like:

ZFS1: (home office / desktop farm)

Norco 4020 chassis (20 drive capacity)
SuperMicro m/b with onboard LSI1068E SAS
HP SAS expander
SAS expander connects to onboard SAS ports and to Norco SAS backplane
SAS expander has external SAS port allowing future expansion to another SAS 
JBOD chassis (e.g. Norco #2 or whatever)
LSI 9200-8e SAS2 controller
External 8 drive SANS digital SAS enclosure
SANS digital connected to LSI9200-8e via two SAS wide ports

9 Hitachi 2TB in Norco case in RaidZ2, plus 2 spares, 12TB usable
I'll be doubling this to 24TB shortly as almost out of space.

Current ZPOOL look like this:



rz2pool                 12T  
 900G   1.2T    42%    /rz2pool
rz2pool/bootimages      12T    36G  
 1.2T     3%    /rz2pool/bootimages
rz2pool/cxxxx           12T  
 2.6T   1.2T    68%    /rz2pool/cxxxx
rz2pool/exp             12T  
 7.3G   1.2T     1%    /rz2pool/exp
rz2pool/mace            12T  
 7.5T   1.2T    86%    /rz2pool/mace
rz2pool/macedb          12T    19G  
 1.2T     2%    /rz2pool/macedb
rz2pool/projects        12T   695K  
 1.2T     1%    /rz2pool/projects
rz2pool/tftpboot        12T   8.7G  
 1.2T     1%    /rz2pool/tftpboot
rz2pool/vmware          12T    48K  
 1.2T     1%    /rz2pool/vmware



The bulk of the data to be backup is from the MACE pool which contains 
real-time traffic captures from a bank of 100 PCs that are doing generalized 
web surfing.  There is about 100GB/week of data added here.


3 Hitachi 2TB drives in SANS Digital case in RAID0 for Amanda holding disk
20+ Hitachi drives for backup "tapes" to be swapped in and out of 5 remaining 
SANs digital "slots"

My overall plan here is to use the Norco case for capacity expansion.
Use 3-drives in the SAS digital for Amanda holding disk as follows:


# df -h
amandadump             4.0T  
 3.2T   785G    81%    /amandadump

# zpool status amandadump
  pool: amandadump
 state: ONLINE

        NAME        STATE    
 READ WRITE CKSUM
        amandadump  ONLINE      
 0     0     0
          c0t21d0   ONLINE    
   0     0     0
          c0t22d0   ONLINE    
   0     0     0
          c0t23d0   ONLINE    
   0     0     0




And then use the other 5 available "slots" in the SANs digital chassis for 
hot-swapping disk-based "tapes" as needed.

I've been formatting individual 1.5TB or 2TB drives to be "tapes", e.g.:


# zpool status tap0102
  pool: tap0102
 state: ONLINE

config:

        NAME        STATE    
 READ WRITE CKSUM
        tap0102     ONLINE    
   0     0     0
          c0t26d0   ONLINE    
   0     0     

# df -h | grep tap
tap0102                1.3T  
 1.3T     0K   100%    /tap0102



Here are some of the challenges/questions:

My full backups are huge (multi-TB) so filling tapes is not a problem.  
However, daily incremental might only be 20-30GB.  Since I'm using 1.5 and/or 
2TB drive for "tapes" I'm able to get good "filling" of the tapes when doing 
full backups, but I'll potentially only fill up 50GB other times....this 
potentially leaves a lot of "tapes' mostly empty and is kinda a waste of an 
expensive resource.

I can work around this by disabling auto-flush...doing a bunch of backups over 
the course of several days until I have at least 2TB on the holding disk, then 
mounting a "tape" and doing a flush...problem is then it requires me to manual 
monitor it which I'd like to avoid...


I've read all of the discussions on using vtapes, but they all center on using 
a fix mounted drive (or multiple drives in RAID).  So I certainly could create 
200 10GB vtapes on a 2TB drive, but I really need a total of 20TB of backup 
capacitiy...how do I setup vtapes on 20 different drives, but I only have the 
"slot" capacity to have 5 of them physically mounted at any given time.  
Bottom-line, I'm unclear on how I could use a vtape approach with MULTIPLE 
physical disks.   I don't want to do RAID as I want the ability to migrate some 
of these backups offsite...doing vtapes on RAID would seem to signficantly 
complicate that process.

Thoughts?
[/code]

+----------------------------------------------------------------------
|This was sent by [email protected] via Backup Central.
|Forward SPAM to [email protected].
+----------------------------------------------------------------------


Reply via email to