copying filesystem

[Kishore Gowda]

Hi all,

We have Oracle 11i datafiles that currently resides on Seeks storage array (FWD SCSI / RAID5); but are planning to migrate that over to Symmetrix frame (Fiber - RAID 10). The host is V2500 running HP-UX 11.0 with connections to both Seeks & Symmetrix. When I ran a couple tests to copy filesystems (using tar and cpio) between these two units, the results were less than encouraging. To copy 27Gb from one fs to another, it took about 2 hours! But the network based backup (Legato Networker) takes about the same time (2 hrs) to backup the entire database (80 Gb!). I am a little baffled as to why is network based backup is nearly three times faster than the local filesystem copy? Any thoughts on this is greatly appreciated.

Regards,
Kishore

[Ross Zubritski]

Kishore,

It soulds to me like the "Seeks" is just plain slow. I am sure the bottleneck is not on the Symm side.

Regards,

RZ

[Kishore Gowda]

Ross,

Thanks for the reply; my concern here is network based backups is pulling data from the same Seeks Unit in 2 hours, whereas to copy data from Seeks to Symmetrix takes three times as long?

Kishore

[pap]

Hi Kishore,

The thing is that the network backup software ( Legato in your case) uses different methodology for the backups. initially when you try to start backup using the software it evaluates the data on the system to be backed up and compresses accoridn g to the backup software configured. Once it compresses the data the amount of data to be backed up reduces to significant extend in short if you want to backup 30 GB of data using the backup software , it compresses almost to 20GB or may be less depending upon the compression factor.

Now, the backup for 20GB takes far less time than actually 30GB of data. LAso the backup format is different in both the cases. Its entirely different thing to compare backup software format and tar,cpio format.

Thanks,
-pap

[A. Clay Stephenson]

Well this is not really a fair comparison. THe backup is 1) multi-threaded - many reader processes are running simultaneously. 2) The write operation to tape does not have to wait for synchronization.

Your tar's and cpio's are actually doing both a backup AND a restore - restores are always slower. Each time a file is copied, a directory must be locked briefly so that no other processes can clobber it - the synchronization part. Also, the entire directory must be read and written for each file copied. While these reads and writes may be in cache,a large overhead is nonetheless added to the restore process.

[Leif Halvarsson_2]

Hi,
What kind of data do you used when testing tar and cpio, if there was a large number of small files you should hot expect high performance.
Tar and cpio is not very fast, there is faster alternatives in HP-UX (ftio and fbackup) but they can't be used for direct copying files between filesystems.
In your case I think the best alternative is to use Networker for backing up the old volumes and restore to the new.

[Kishore Gowda]

Yes, I am seeing a better throughput with copying large datafiles (25Gb/hour) as apposed to a ton of small binary files (10Gb/hour). It didn't occur to me that backup software is multi-threaded and uses different compression method, which can be lot faster than tar or cpio. Thanks again for all your replies.

Kishore

[Michael Steele_2]

Sounds like legatto is multi threaded and using multiple backup jobs all running concurrently to one or more tapes. Verify this with your legatto backup administrator.