Unsolved

This post is more than 5 years old

7 Posts

686

May 30th, 2008 01:00

Poor performance cloning from VTL to physical LTO3 - NW733, Solaris 9

We have NetWorker 7.3.3 running on Solaris 9 on a beefy Sun V440. I backup to VTL and then make a clone to take off-site. My problem is that clone speed is about half the speed it should be which seriously impacts my backup windows.

The VTL is connected with FC and the LTO3 drives with U320 SCSI.

Even though I can read data from the VTL at over 140MB/s and write to LTO3 tapes at 105MB/s I can only get a maximum of 69MB/s for cloning speed. I have done speed tests using dd, save and recover which all match exactly.

Before I implemented the VTL I used the same disks as an AFT device and I could clone or stage from AFT to LTO3 at anywhere from 100MB/s to 150MB/s (depending on the compressibility of the data). So clearly there is no underlying hardware issue.

I suspect that the tape subsystem in Solaris 9 is to blame here as the only thing that has changed is that Solaris now treats the VTL disks as a tape rather than a disk now. It is interesting that my clone speed is almost exactly half the speed of the VTL as if it is getting a block of data from the VTL, then sending it to tape and then getting another block from the VTL and so on.

Has anyone else seen this? If so will upgrading to Solaris 10 fix it?

6 Operator

 • 

14.4K Posts

 • 

56.2K Points

May 30th, 2008 05:00

This could be easily due to multiplexing. Even your are using VTL, multiplexing is emulated as well. What you could do is simply do one stream backup and try to clone and stage that after manually to see what speed do you get. If it is the same then it would be probably something within subsystem (unless they share same HBA).

7 Posts

May 30th, 2008 07:00

Great idea. I've had this issue for three months and had not thought of that! It makes sense because most of my VTL drives are running two sessions! My backups are running now so I'll have to wait until next week to test. I'll report back what I find! I hope this is the solution!

Thanks,
Rob

7 Posts

May 31st, 2008 06:00

Hi Hrvoje,

Did some testing today. It is definitely NOT caused by multiplexing. Did some tests with dd and with nsrclone and got the same results. About 69MB/s when I should be getting a minimum of 100MB/s (more with compression on LTO3).

Brings me back to thinking that is is caused by the tape driver architecture in Solaris 9.

Thanks,
Robert

6 Operator

 • 

14.4K Posts

 • 

56.2K Points

June 1st, 2008 23:00

The only way to verify that is to zone that same drive you used for test to some other system (with same HBA setup/sharing) and to see what speed do you get.
No Events found!

Top