> On 7 Feb 2018, at 21:26, Zoltan Forray <zforray AT VCU DOT EDU> wrote:
> 
> As you recall, we have been trying to figure out an alternative method to
> backing up DFS mounted ISILON storage since the current method of 80+
> separate nodes accessed via the Web interface of the BA client is going
> away.  Plus the backups are taking soooooo long, we have to determine a
> better way.
> 
> So, doing some digging, one solution that seems to be touted is using
> NDMP.
> 
> We have absolutely zero experience with NDMP  and are looking for some
> guidance / cookbook / real-world experiences on how we would use NDMP to
> backup ISILON storage (>400TB and hundreds of millions of files) and make
> it accessible so someone from a help-desk like environment could handle
> file-level restores!

I don’t like TSM NDMP one bit, and I guess it’s no worse than any of the other 
backup vendors’ implementations, because NDMP is just what it is, and that is 
not much. I would seriously look into some form of async replication native to 
ISILON, something like netapp's snapvault, first. Yes that requires a a huge 
pile of disk just for backup, but it will probably be worth it. Even if the 
investment is quite high. Don’t forget with TSM terabyte licenses you’ll be 
paying a lot (a huge lot!) to IBM for your NDMP backups.

You can basically NDMP via LAN and via SAN. The latter has the disadvantage 
that the TSM server running the backups must be the library manager for those 
tape drives. I would have loved to see that IBM would make NDMP and Library 
Managers play nice, but alas… NDMP via LAN allows you to use normal disk and 
tape based storage pools, via SAN you’ll need to create a separate tape pool in 
the right format (ndmpdump). Also, you can’t run copy storage pool on those is 
you use SAN. On 8.1.2 and higher (if you dare go there) you could even use 
directory containers.

The current customer has NAS systems which share directories (called virtual 
volumes) rather than separate file systems. To be able to make a more granular 
backup/restore they use virtualfsmappings in TSM. This works surprisingly well. 
Now a huge NAS file system becomes (usually) a far more manageable directory. 
So not 200 TB in one huge lump to backup, but mostly directories of under 1 TB. 
The backups are slow, but on average manageable. We have a few exceptions that 
we backup via the share because they are just too big to manage via NDMP. 
Problem with NDMP is that if (with TSM 8.1) a single transaction spans more 
than 90% of the active log, the transaction gets killed by TSM. This is on 
average a good thing, but that makes the combination of a busy TSM server with 
loads of files and NDMP not a happy one, at least not for those few huge 
virtual volumes.

So basically:

- look at other solutions (snapvault or whatever it’s called for your NAS)
- then again look at those solutions
- virtualfsmappings might make things more manageable if you decide to go with 
TSM anyway
- SAN and LAN both have disadvantages, neither one is perfect
- maybe a dedicated TSM instance to avoid issues with long running ndmp dumps

> 
> Or if NDMP is the wrong direction, please tell us so.
> --
> *Zoltan Forray*
> Spectrum Protect (p.k.a. TSM) Software & Hardware Administrator
> Xymon Monitor Administrator
> VMware Administrator
> Virginia Commonwealth University
> UCC/Office of Technology Services
> www.ucc.vcu.edu
> zforray AT vcu DOT edu - 804-828-4807
> Don't be a phishing victim - VCU and other reputable organizations will
> never use email to request that you reply with your password, social
> security number or confidential personal information. For more details
> visit http://phishing.vcu.edu/

-- 

 Met vriendelijke groeten/Kind Regards,

Remco Post
r.post AT plcs DOT nl
+31 6 248 21 622