LTO tape speeds

What are the best ways to calculate LTO write speed?

I recently went from writing to LTO 6 from a single 3TB 5400 rpm drive to 3x3TB 5400 rpm drives in RAID zero and want to test the speed difference. I don't think it is a much as I was hoping for but I'm not sure of the best way to test it other than with a stopwatch, which isn't practical.

I'm transferring hundreds of thousand of .dpx files and even with the 9TB RAID my LTO drive still slows down and rewinds ("shoeshining"?) occasionally.

2 Answers

The maximum LTO tape drive speed is well documented in Wikipedia, for LTO 6 it is 160 MB/s. The minimum speed for Tandberg LTO-6 drive is 54 MB/s (tandberg.com). Others are similar. An 5400 RPM hard drive is faster than than 54 MB/s, but only in case of sequential read. You have many small files, so your read pattern is more like random reads. That means a much much lower actual hard disk read performance. In such cases it is better to configure a several gigabytes long buffer on a second hard disk. This caches several small files together. For example the Bacula/Bareos backup programs have a feature for this. This minimalizes shoeshining if your network/disk performance is not enough to feed the tape.

Essential is the minimal drive speed which is 54 MB/s for uncompressed data for HP drives. (afaik HP produces for Tandberg)

You can also try to disable the compression. This can be done with mt-stVerify that you run mt-st with mt --version.

# mt -f /dev/st0 compression 0 # off
# mt -f /dev/st0 compression 1 # on

You can use mbuffer to analyze your problem. The commandline would be something like

dump -0u -f - /data/ | mbuffer -R 80M -P 90 -m 4G > /dev/nst0 

if you have some data to backup in /data. I use a very fast SSD and still have shoeshining effects with LOT-6 which I do not understand yet and need more tests. Perhaps it is rewinding and not shoeshining? (see How often does an LTO tape brake and return for a full write?)