IBM LTO (Ultrium) 2 on Solaris

[julianbarnett]

We have recently taken delivery of an ADIC Scalar i2000 with 8 fiber attached LTO2 drives running on Solaris 8 with the latest st driver patches.

I accept that with 4 drives per HBA we will not be able to get the maximum throughput to the tapes due to bandwidth limitations. However, even with a number of savesets from EMC attached discs streaming to a single drive, I rarely see the Legato throughput peak above 20Mbs. Much of the data on the discs is already compressed but I should have thought I would get closer to 30Mbs.

I notice the default blocksize for the LTO 2 tapes in Legato is 64Kb compared to 128Kb for our old DLT7000s. I assume these values are set in the OS/Legato driver since we have never tuned these, and wonder if increasing it using the Legato initialisation variables would have any significant benefit. If I do increase this blocksize will I still be able to read data off the LTO tapes we have already written to using a smaller blocksize? - I believe the blocksize is written to the head of the tape when it is labelled by Legato.

I would like to hear if anyone has seen anywhere near the claimed 35/70 Mb/s throughput using a similar setup.

Regards Julian

 

[RichS24]

Hi,

the problem with tape speed issues is always going to be finding where the bottleneck is. With your particular set up, the first thing that has to change is the number of drives per hba. 4 DLT7000 drives on one hba would kill it, let alone LTO's.
Secondly, you need to establish the speed at which the emc disks are passing data through to the drive. Try doing a backup of a large file through to /dev/null and see how long it takes. This will then let you know if the disks are the problem.
If the disks aren't the problem, then you need to start experimenting with the hba. See if you can get a standalone drive on loan, ensure it is the only drive plugged in to the hba and try a test backup.

In answer to your question. Yes, i have seen speeds of up to 50mb/s on LTO Ultrium 2's, but we never have more than 2 drives per hba.

I hope this helps.

Cheers

Rich

Cheers
richs24

 

[wallace88]

Are you using sortware compression during backups of your clients? In some cumstances, usong software compression slows down the data transfer rate, and therefore the writing speed.

 

[cyberted]

make a test with bigasm (with and without compression)


To set up a bigasm test:
1. Create a file, bigasm.file, that contains the following command:
bigasm -Ssize : filename
For example:
bigasm -S100M : bigfile
2. Save the file using the NetWorker server. For example, if the directive file
is called bigasm.file and the NetWorker server is called jupiter, enter the
following command:
save -s jupiter -f /bigasm.file /tmp/filename
The amount of time it takes for the backup to complete indicates the
efficiency of the SCSI throughput and the write speed of the device.
The bigasm module can also be used in a scheduled backup by creating a
directive for it in the nwadmin program.


watch your disk I/O with sar -d and your cpu idle timer.
(post them)

we have LTO1 with an EMC Symmetrix and I only see 12-17M/s with compression (factor 3/1). So if I write 17M/s to the tape I read 51M/s from your Symmetrix. You would not get more throughput.


cyberted

 

[cyberted]

test your backup with parallel streams this will increase performance.



cyberted