tmac,

 

The tape drives are fiber channel, and the connectivity all goes through a cisco MDS san switch. We have 6 tapes in the system, and normally run 4 ndmp backupin parallel. The path to tape should be a 2gbps FC link. The back end infrastructure obviously was built to sustain a much higher bandwidth.

 

I'll try to run the null device dump and come back with the results

Thanks!

 

On 12/18/06, tmac <tmacmd@gmail.com> wrote:

try dumping to a null device instead of tape to test raw speed without tape.

pick one of your dumps, and then do something like:
dump 0f null /vol/vol0

You can see how long the dump takes and compare it to tape.

How many simultaneous dumps do you do? are the tapes on FC or SCSI or
over the net?
how fast is the path to tape?

if it is over the net, did you check for any errors, like using netstat -a?



On 12/18/06, Ammar Zolkipli <ammar.zolkipli.lists@gmail.com> wrote:
> Gents,
>
> Let me first explain the overall layout. I've got a D2D2T design. Both
> primary and secondary filers are v-series, and snapvaulted. We have 16TB of
> CIFS home dir data volumes backing up to tape via NDMP. And the tape backup
> is on LTO-3 running at sustained throughput of 20mbps, (F**king slow, I
> know).
>
> Does anybody have experience how to speed up NDMP backup? At this point, the
> LTO-3 is shoe-shining as the minimum sustained throughput on that drive is
> 40mbps. And also, does anybody have any experience with NDMP to tell us why
> this is happening?
>
> Thanks


--
--tmac