Check disk and CPU usage on DB client.

I guess disk busy rate reached to 100% or so.

2 * 8GB FC connection to storage will not do the job.

1GB=100MB/sec

8GB=800MB/sec*2*3600 (one hour) =  5,76TB per hour * 6 hour = 34.56TB

I think you only got capacity to backup half of the target of 60TB.

I doubt that the storage Oracle sit on has enough performance rather than link speed.

Approximately 3GB/s is needed. Well configured storage and well layouted oracle may easily achieve this, but should check.

You see that the more tape drives you write to the slower it gets - i have seen this a lot when systems are overstretched and the bptm process has multiple instances running.

As Nicolai says you cannot actually do it anyway based on the required throughput to hit a 6 hour window.

So to get 60TB in 6 hours means 10 TB per hour .. so 10240 GB per hour .. so 170.6 GB per min .. so 2.8 GB per second .. so 2913 MB per second.

This would need 4 x 8GB fibre for the read and for the write (so 8 in total) and tape drives that could actually write fast enough in total.

But even then this would be driving the system totally flat out and i doubt you could drag it off the disk that fast.

Take a look at nbperftest which will answer that question for you.

Either way i doubt that bptm will cope without massive amounts of RAM and O/S and BUFFER tuning.

Maybe better writing to disk to get more out of it?

To protect such a large database you are better off using disk snapshot technology.

Restore of 60TB via tape will take a very long time. And just a single tape drive in a marginal health state will prolong the restore into days.

llgiz:Yes I know this may be sad new for you. But we are really trying to help you. You asked for a solution, but has equipment already been purchased ?

An interesting problem to have but the whole point of design is to not have issue like this...anyways, what are you using :agent backup , flat file, or....?

What are your data sat on...vx,nfs,ufs,zfs?

What storage are you using? 

There might be  a chance to get some better throughput (first things first you have tuned your tape buffers right? It looks like it) However this wont scale as the pipe for drives is only 2x8gbit which might be good for maybe 3 or 4 lto5 peak.

Anyone ever zoned in tape drive and storage on the same fibre to leverage duplex? Someone with more fibre expertise might be able to comment. If that works things could improve as you pull data down one path on a fibre and push it out on the other fibre path.

Jim

Dear All

We have next schema:

4 port 8 Gb/s FC card connect via SAN to Storage.

4 port 8 Gb/s connect to SL3000.

24 dirve from SL 3000 connect to server (6 drive to 1 FC port).

Afret start backup to 12 drive I see I/O 96% and act_wait 250 from OS, but from storage site 20% I/O busy.

Maybe problem with Server I/O bus ?

Ilgiz.