How to View Tape Drive Performance..

[Alzhy]

On Solaris systems, iostat -xcn shows not only disk LUNS but tape LUNS as well - where current performance in blocks (KB) per second are displayed.

Any equivalent or tool that could do this on HP-UX11i?

Thanks!

[A. Clay Stephenson]

In HP-UX, the best I/O Metric for tape drives that I have found is Glance -> Process List -> Open Files and then observe the change in the File Offset as a function of time.

[Shahul]

Hi,

I am not sure of any command which give tape performance in HPUX. One way to calculate performance of tape drive is to verify a successfull backup or by using Glance plus. Again the performance varies according to the tool which you are using for backing up. Normally fbackup gives better throughput.

Good luck
Shahul

[Alzhy]

Thanks A. Clay.. My goal is to actually compare (and find out what is wrong) between a Sun attached Library and an HPUX one (same model and drive type,DLT7K and same connectivity - SAN bridged). On Solaris environment, iostat effectively portrays performance of all my tape LUNs and the same DLT7K's stream (HW, compressed) at no less than 6 MB/sec with peaks of 9-10 .. which is the rating for DLT7K drives..) On HPUX environments, we are only getting a measly 4-5 MB/sec even if we have configured the device to do compression.. And the only way I have measure so far is to backup a fast filesystem to a tape drive and monitor via iostat how that drive on which the filesystem resides is actually sucked...

Has anyone written C code (via whatever API stape interfaces with) that could effectively portray tape drive performance/throughput?

It would be nice for HP to enhance iostat so it also shows tape LUNs performance..

[Rodney Hills]

If these are DAT drives you can use the program "tapeinfo". It can tell you how many bytes go to the tape drive and how much if compressed.

You can find tapeinfo from hp's support web site.

HTH

-- Rod Hills

[Alzhy]

Unfortunately these are DLT7000 drives.. I will look at tapeinfo.. There should be no differences inthe way a tool would extract performance data between a DDS/DAT and DLT/LTO/Ultrium/AIT/SAIT...

[Tim Sanko]

I don't trust all the tools out there in the world. I make my own benchmarks, as several vendors numbers weren't real.

There are some consistancies that you can piggyback within Solaris & HP.

There are also several ways to do this scientifically.

Method 1.

First collect the data that you want to model. (5-10% of similar files to what you will back up).

Second use a common utility (cpio or tar) to get a comparison.

Third get out your calculator.

Get the raw numbers that will be relevant including tape changes and plot them...

Method 2.

If you don't have glance, you do have netstat,sar,and lanadmin. If you are using IP you can calculate the throughput using netstat, the disk throughput will parallel tape volume when measured with sar. In this manner sar is much more usefull than measureware. If you don't backup the drive sar output is on, it is easy enough to dump into a spreadsheet. and extrapolate the IO.

Subtract the sar IO from the sar output for the period of the benchmark, and you are in business.

If this is a production box that has similar performance patterns and you cannot take it for the benchmark tests. You can get benchmark without backup, and benchmark with backup...

We currently have 50 GB of sar sara for several machines. Spotting problems via activity level is not trivial, but has given us the ability not to have to upgrade the production server. (additional work for SA on salary vs $500,000.00 server replacements)
This is a "no brainer" for our management...

Anything that is reproducable, by period gives you a solid baseline. The baseline for Tuesday during monthly close is heavier than Tuesday of non-close week. by 15-25 % CPU utilization, but 40% disk utilization.
Tuesday of Quarter close is 5% heavier in terms of disk IO and CPU use.

Sar can help you find the hot controllers, drives etc...

If you use EMC I recommend PowerPath...

Tim

[Alzhy]

Tim, that's what exactly I am doing.. using the available tools to at least portray how my clients are feeding backup data to my backup server, how disks are feeding data to the drives...

I am just sort of "surprised" HP's iostat implementation only includes disk devices.. IOstat should be able portray all or most IO.. tty, disk and tape (would you believe even NetWork on earlier versions of Solaris as an "undocumented" parameter)...

Thanks any way .. your points will be educational to the other ITRC members...

[Tim Sanko]

I have found discrepancies in many tools. sar to measureware in particular... I trust sar more than most other tools.

I still run benchmark tests after upgrading sar/measureware etc...

It has been around a long time and has stabilized in the last 28 years that I have been doing UNIX work...

[Steven E. Protter]

I'm attaching a script that will measure this performance and other metrics for you.

Good Luck,

SEP

[A. Clay Stephenson]

One of the challenges to evaluating the performance of a high-performance tape drive is trying to feed it especially with a single-threaded process. You will get the best performance using fbackup with multiple readers or even better OB2/DP. Using OB2, even with an old D-box, I routinely get a sustained rate of 6MB/s and peaking to 9-10MB/s depending upon the compressability of the data stream --- and this is over copper SCSI. To measure this, I use Glance to look at the OB2 process list and then monitor the file offsets.

You will have a difficult time trying to measure this with a tool like dd. Above all, don't be tempted to use a seemingly good command like
"timex dd if=/dev/zero bs=256k count=10000 of=/dev/rmt/3m". As yourself, how would an infinite stream of NUL's compress. You might have better luck using if=/dev/urandom but even there you may be more limited by the speed of the RNG rather than the tape drive.

[Alzhy]

Thanks for the replies so far.. your views are indeed helpful... but interistingly enough, the backup tools offer drive status/monitoring already.. I was just hoping there would be a ready tool ie. iostat or tapestat? that'll validate what my backup software is reporting or that I could use to measure throughput when using say vxdump, tar, cpio etc..