Slow NDMP recover from diskbackup device

We have done more testing and removed the DIRECT=y Application information in the Client configuration in Networker

The settings is now: HIST=y and UPDATE=y

And now the restore speed is sufficient.

Anyone who have seen this before?

DIRECT was requiredin the past when restoring files only (no directory is marked for restore).  I haven't seen this before, but it is possible that this option might not be something to be used with disk devices.  If you have some tape device, you can easily test it.

Thanks for the reply Hrvoje.

Let me explain more about our setup. We have been running our NDMP backup directly to LTO attached tape drives for several years. But have decieded to use a disk device instead and use 10Gbit ethernet as transport to backup our filers by using NDMP.

Have also thinking about using NFS, but there is issue by using this method also.

So what recommendations do you have when it comes to spesifying the NDMP Application information spesific in the Networker client. A I have stated before we have always used these settings:

HIST=y
UPDATE=y
DIRECT=y

We have also tried using NSR_NDMP_DDAR=y; export NSR_NDMP_DDAR before we ran recover/nwrecover to be sure that DAR worked.

Have also tried

NSR_NAMEONLY=No
NSR_RENAMES=Yes

To fix the problem 3095855 (missing files if directory moved)

So if you or someone else have any input's it is greatly appreciated.

rgds

Karl Dag

Application information should contain what you find fit.  I usually check filer's documentation for available options and use those which are required for specific environment only.

On the separate note, did you do any tests which went to something else than zfs?  I know in the past some very strange things could be observed with zfs (especially if dedup was on).

I have not tested on anything else than zfs. We was thinking about using UFS but deceided not to use it after the tests we ran from Solaris.

I now that there was strange tings that could happen in the early days ZFS but it has matured a lot since then. And dedup is not an option in Solaris 10. Oracle has removed the option, probably due to Solaris 11.

After some tests I have found a setting that give us about 200MB/s pr saveset backup speed and about 110MB/s restore speed.

The settings is as follows

Networker:

     In Application information:

          HIST=y
          UPDATE=y

     Before running nwrecover or recover. Set the envronment varable: NSR_NDMP_DDAR=y

Netapp (Ontap version 8.0.1)

     ndmpd.access                 all       
     ndmpd.authtype               challenge 
     ndmpd.connectlog.enabled     on        
     ndmpd.data_port_range        all       
     ndmpd.enable                 on        
     ndmpd.ignore_ctime.enabled   off       
     ndmpd.offset_map.enable      on        
     ndmpd.password_length        16        
     ndmpd.preferred_interface    e1b        (value might be overwritten in takeover)
     ndmpd.tcpnodelay.enable      off

I will also open a case with EMC to see if they have any comments.

Ah, that's NetApp 8... no experience with that one yet.  Unless I'm mistaken, 7.6SP2 has been qualified with NetApp8... not sure about other versions.

HI all

Trying to restore 500KB file from full backup of 10TB...looks like reading entire full backup...is that the behaviour? if so, it would take ages just to restore small file..

Application Info

HIST=Y

UPDATE=Y

OPTIONS=NT

SNAPSURE=N

NSR_NDMP_TMP_DIR=T:\NDMPTmp

Celerra 7.0.35-3

Thierry,

New problem, new thread please.  And include information if that is DSA based backup or not.