DD Performance

gernthefish

ADSM.ORG Member
Joined
Mar 3, 2005
Messages
139
Reaction score
0
Points
0
PREDATAR Control23

We are replacing our "disk to tape" solution with a DD4200 and are noticing DD backups/restores are slower. Of course, this is not going over well since disk is supposed to out-perform tape. Any thoughts on this? Can we continue to use the local TSM server drives and set up the DD as a NEXT stg pool?
 
PREDATAR Control23

I'd open a support call with DD if you are not getting the throughput you are expecting.
Are you connecting to the DD via the network?
How many GB/hr do you need to transfer?
Is your connection(s) fast enough for the data transfer you need?

I don't need the answer to those questions, but they are the questions you need to think about. Have to keep in mind that TSM can't send data faster than slowest link, and DD can't write data faster than it receives it.
 
PREDATAR Control23

How is the DD device attached to the TSM server?

DD's are normally NFS mounted to the TSM server and defined as devclass=file. I have gone through several DD devices from DD880 to DD990 and the speed is amazing while transferring with multiple streams. I get transfers like 600 MB/s in two NIC paths or total of 1.2 GB/s with 80 to 100 simultaneous streams (from multiple nodes).

Check your NFS setup. Mine is ethernet-over-fiber at 10 GB per line.

I am betting there is nothing wrong with the DD. This looks like a network setup from the TSM side or how the TSM server is linked with the DD device.

On another thought, did you define the DD as VTL? If so, how many drives have you defined? Have you setup for the right LTO emulation? Have you setup for the right LTO capacity?
 
PREDATAR Control23

Hello
With dd860 and TSM (CIFS) , mtree show performance are slow 250MiB/s (read) when tsm backup on copy pool (4 process) , i think this is because of the high compression (80%) and I found something else,
RMAN client has an NFS share, on the same mtree
the best way to use the TDP agents and delete the NFS share

With agent the flow is between 400 and 500Mibs / s
 
PREDATAR Control23

Hello
With dd860 and TSM (CIFS) , mtree show performance are slow 250MiB/s (read) when tsm backup on copy pool (4 process) , i think this is because of the high compression (80%) and I found something else,
RMAN client has an NFS share, on the same mtree
the best way to use the TDP agents and delete the NFS share

With agent the flow is between 400 and 500Mibs / s

Two things strike me here:

One:

You mentioned compression. Are you compressing data coming in from the node? If so, that is where your problem lies. Node data shoudl not be compressed. If compressed, DEDUP will not work on the Data Domain.

Two:

Are you using a Windows based TSM server? If yes, check your CIFS and network setup. Also, the NFS mount to the DD860 is OK IF Oracle is backing up to the DD860 directly and uses DD Boost. This does not need the TDP portion of TSM which, in most cases, runs backups faster.
 
PREDATAR Control23

Sorry i want to say dedup, the cifs parameter it s like datadomain best practice, in my case dba make many cross check after full backup and the datadomain have 20 000 ops, i think its to much
 
PREDATAR Control23

Sorry i want to say dedup, the cifs parameter it s like datadomain best practice, in my case dba make many cross check after full backup and the datadomain have 20 000 ops, i think its to much

DEDUP rate has nothing to do with ingest rate on the Data Domain.

The ingest rate depends on the speed of your network, the TSM server I/O handling capacity and the amount of allowed nodes that simultaneously access the TSM server.

Check for maximum 'mount limit' defined for the devclass that points to the Data Domain DD860 CIFS connection.
 
PREDATAR Control23

On windows server , without dd boost on rman client

I just want to add that someone had the bad idea to put the cifs and nfs sharing the same mtree ,
 
Top