I forgot to mention that the files on the NetApp that are being backed up are 99% large files like 10+GB in size each so the slowest is not due to a lot small files in case anyone is thinking that route.
Sounds as NIC/port setting which is messing up your connection. People using NetApp usually like to mention jumbo frames as well - while those will make difference you will need special network hardware to support it too (nevertheless, your speed should be arond 30-40MB/s without it with this filer I believe).
Well, it's a little overdue I guess, but I'm awfully sure you're using de-dup on your FAS. I've seen horrible NDMP performance from those boxes with both Networker and NetBackup combined with de-dup, so, NW is not to blame.
Try it with a volume without de-dup. You can also try ndmpcopy for testing.
That may be a solution indeed if you're using a dedicated backup (V)LAN, and you're accidentally backing up via the production/front LAN.
If you're doing local backups (via an rd=netapp:/nrst0a or something like that, so a direct volume to tape without LAN traffic) this will make no difference.
Just to add to Sanjay's advice.....if the data is being written to a device on a NetWorker Storage Node rather than the NetWorker server itself you should also add the -P option and make the value of -P the hostname or IP address of the Storage Node.
you should update your ONTAP to 7.3.1 or later, this should fix it!
Turns out that earlier ONTAP versions cannot handle deduped filesystems correctly and slow down horribly in cases of VMware VMDK disks and NDMP backups (actual problem lies in sequential reads of zeroed out files).
I'm getting at best 8MB/s backing up my NetApp 3020 via NDMP and this seems very slow to me. The tape to tape clone runs at 50+MB/s. Non-NDMP backup of servers runs at 90+MB/s.
I have gigabit connection across the entire network and the NetApp unit is on the same subnet at the Networker server. I'm using the backup command "nsmdmp_save -T dump -M" on the NDMP enable clients.
He needs -M if he's doing backup over network. In that case you might use -P as well to force which interface to use (in case that storage node is multi homed).
He needs -M if he's doing backup over network. In that case you might use -P as well to force which interface to use (in case that storage node is multi homed).
I stand corrected :-0). We use brocade switch (lan-free) backup and get a average throughput of 120MB/sec. (Sometimes peak is 150 MB/sec)
However, if non-NDMP backups are fast, then probably it's not a network issue.
I would recommend a check of ndmp version and other things like de-dup check, like others suggested.
Has any of the suggestions here helped with your performance issue?
Keep in mind that EMC recommends using a GigE connection on both the NetApp and the NetWorker Server (or NetWorker Storage Node) with full duplex enabled on both.
Please let us know if you have made any progress with this issue.
fl1
89 Posts
0
February 25th, 2008 09:00
ble1
4 Operator
•
14.4K Posts
0
February 25th, 2008 12:00
AlexK3
8 Posts
0
January 5th, 2009 07:00
I've seen horrible NDMP performance from those boxes with both Networker and NetBackup combined with de-dup, so, NW is not to blame.
Try it with a volume without de-dup. You can also try ndmpcopy for testing.
Good luck! Let me know if it helped...
AlexK3
8 Posts
0
January 7th, 2009 05:00
If you're doing local backups (via an rd=netapp:/nrst0a or something like that, so a direct volume to tape without LAN traffic) this will make no difference.
rmines1
49 Posts
0
January 8th, 2009 09:00
AlexK3
8 Posts
0
February 27th, 2009 06:00
you should update your ONTAP to 7.3.1 or later, this should fix it!
Turns out that earlier ONTAP versions cannot handle deduped filesystems correctly and slow down horribly in cases of VMware VMDK disks and NDMP backups (actual problem lies in sequential reads of zeroed out files).
See http://episteme.arstechnica.com/eve/forums/a/tpc/f/833003030931/m/854006105931
Good luck!
Alex
satishs
50 Posts
0
February 27th, 2009 07:00
via NDMP and this seems very slow to me. The tape to
tape clone runs at 50+MB/s. Non-NDMP backup of
servers runs at 90+MB/s.
I have gigabit connection across the entire network
and the NetApp unit is on the same subnet at the
Networker server. I'm using the backup command
"nsmdmp_save -T dump -M" on the NDMP enable clients.
Am I missing something somewhere?
--------------------------------------------------------------------
We use the command
"nsrndmp_save -T dump"
and NDMP backups work correctly. Try removing the "-M"
ble1
4 Operator
•
14.4K Posts
0
March 3rd, 2009 09:00
satishs
50 Posts
0
March 3rd, 2009 10:00
that case you might use -P as well to force which
interface to use (in case that storage node is multi
homed).
I stand corrected :-0). We use brocade switch (lan-free) backup and get a average throughput of 120MB/sec. (Sometimes peak is 150 MB/sec)
However, if non-NDMP backups are fast, then probably it's not a network issue.
I would recommend a check of ndmp version and other things like de-dup check, like others suggested.
rmines1
49 Posts
0
March 13th, 2009 15:00
Keep in mind that EMC recommends using a GigE connection on both the NetApp and the NetWorker Server (or NetWorker Storage Node) with full duplex enabled on both.
Please let us know if you have made any progress with this issue.
Regards,
Rita Mines