ADSM sets it's own tape blocksize in the code.  It doesn't matter what
block size you give it with SMIT.  The block size you set in SMIT will
only be  in effective for tar/dd/etc., but not an application using
IOCTLs to configure the device.  ADSM sets the blocksize to Zero (ie.
variable) and blocks are generally written in 256K buffers until the
last block in a transaction, which would generally be less than 256K.

As for performance, there are a number of things to speed it up, but
also keep in mind the bottlenecks in your environment, including LAN
speeds, disk speeds (a lot of disks are slower than a streaming
3590), and ADSM's database updates.  Tar and dd don't have to deal with
most of that.

Look at the the PERFORMANCE TUNING GUIDE V2.R1 (SH26-4034) for
guidance.  Some of the server options that tune performance are:

BUFPOOLSIZE
EXPINTERVAL
LOGPOOLSIZE
MAXSESSIONS
MOVEBATCHSIZE/MOVESIZETHRESH
TXNGROUPMAX

As of ptf12, there's also USELARGEBUFFERS that is a storagpool parm.

The Client options COMPRESSION, DIRMC, QUIET, SLOWINCREMENTAL, and
TNXBYTELIMIT also have a performance impact.

Craig Bell