Some time ago I built an Amanda 2.5.x system on SuSE 10.3 (using the repo version) with an IBM Ultra 160 scsi LTO3 and was getting amtapetype figures of around 65000 kp/s.

I recently built a second system on Suse 11.1, again using the repo version 2.5.2.1 on an IBM x3550 with a Quantum SAS LTO4 connected via an LSI SAS3801E. This only gave me ~45000 kp/s.

Rebuilding the above to Suse 11.2 and the supplied 2.6.1.1-3.4 dropped this figure even more to ~37000 kp/s, a figure confirmed by a third build on a different cpu but the same other components.

I get over 200MB/s writing through the HBA to a SAS raid, and an LTO4 should be good for 120MB/s and is audibly spending a lot of time waiting, so why the pants performance?

Incidentally, what is the relation between the amtapetype "kp/s" and the more usual measure of throughput in MB/s? I presume it stands for kilo packets/sec but I am not sure of the packet size, its clearly not the block size.