Reclamation process - Tape to tape - Quick question

[LarryB]

We have a 3583 TS3500 Library Tape unit. This is not for storage pools on disk. For reclamation using tapes only, it requires 2 tape units, that is, copy from one to the other. Does the data transfer/path exist within the tape unit OR does the data go back to the TSM Server, then back to the destination tape ?

Source TAPE----------------> |
| TSM Server (Solaris)
Destination TAPE <----------- |

 

[max]

hi,
"data go back to tsm server" ? Whereabout ? They will remain inside the unit, tsm server will just receive some signalling with info about the ongoing process.

cheers
max

 

[Efim]

As far as I know...
During reclamation using two tapes data move directly from tape to tape using TSM server hardware (FC-PCI...-FC or SCSI-PCI...-SCSI).

It isn't server-free transmission using DATA-MOVERS and COPY SCSI commands.

Information about objects transfers will be updated in TSM server database.

Efim

 

[BBB]

Just to clarify, and max is wrong by the way:

The data will be read from the source tape, back to the TSM server, and then the TSM server will write the data to the destination tape. So it will cause a lot of SAN traffic if there's lots of data on the tape as the read and then the write will travel across the SAN.

 

[max]

hi bbb,
well i'd like to understand your point because i cannot figure it.
Where should tsm server store those data and most of all why should it do that ?
As far as i'm concerned it mounts one tape into a drive to read its content and it mounts a tape to another drive to write on it, what has the SAN to do about that ? It might even be a scsi connection or a fiber direct connection to host adapters without crossing any san switch.

Can you please explain ?
cheers
max

 

[heada]

BBB is correct. The TSM server will mount two tapes (regardless of where they are, SAN; direct fibre or SCSI) and read from one tape and write to the other. The data is not stored on the TSM server during this transaction but it does flow through the TSM server. All new meta data is stored in the TSM server database (the fact that a copy now exists on the second tape)

If you monitor your SAN while performing a reclamation or a backup stgpool (tape to tape) then you'll see data flowing in and out of the TSM server as well as the tape drives. It is harder to monitor a direct connection or a SCSI connection but you would see the same thing.

-Aaron

 

[cheese]

to summarize:

tape to tape copies run from source-drive over the net/fibre/san to the tsm-server and then to the destination-drive.

if the sourcetape is a full tape (without alot "holes") and only big files (ndmpbackups), which performance should i get?
these are lto4 drives and i think i should get up to 120Mbytes per second. in fact i got around 30 Mbytes (somewhere around 4-5 days for around 12 Tera data).

 

[heada]

Performance depends on several factors and without knowing exactly how everything it setup it is very difficult to give an estimate. 120MB/sec is a good high number but I would call that total throughput and not drive speed. Since the data has to flow through the TSM server, you would see 60MB/s in reads and 60MB/s in writes. Another problem is that unless everything matches up perfectly, the tape drives are going to have to stop occasionally to reposition or wait on the other drive.

-Aaron

 

[cheese]

is this stop and go true for "full" tapes, too. my source-tape is completely full and there should hopefully no stop/go be needed. what makes my confused is that 30MB/s is far away from uncompressed 120MB/s, which my lto4 should be able to provide.
how long does your backup stgpool last for such full tapes?

 

[heada]

It is very rare that a TSM server runs at the exact same speed as a streaming tape drive. Because the TSM server can't keep up with the tape drive, the drive must stop now and then and reposition the tape under the read/write heads.

It is very rare that I perform a backup stgpool on a tape pool that has full tapes and even then, I use 3592 drives and not LTO4 so my numbers wouldn't match yours. Also, 120MB/s is the max the drive can support. It doesn't mean that it is possible or common to run at that speed.

-Aaron

 

[PJ]

We see between 90 and 170 MB/s per LTO4 pair during backup stgpool with large files on AIX/P6 and a rather wide 4GBit SAN with no more than 3 drives per adapter. I'd agree with Aaron that you rarely ever see theoretical speeds with the latest LTO4 and 3592 because the server usually can't keep up - but 30 MB/s strikes me as being far too slow. I'd run an instrumentation trace first to see where the process is wasting its time.

PJ