star tape I/O suggestion
[email protected] Thu, 3 Mar 2011 19:12:23 -0000
| Newsgroups | gmane.comp.archivers.star.user |
|---|---|
| Message-ID | <402b70c8ff0073d763ed3d8b98e6906d.squirrel@ssl-webmail-vh.clara.net> |
Hi, The default blocking factor of star (and other tar programs) is 20, so data is written in chunks of 10240 bytes. While using that small a tape block size allows for wider interchange, writing one block at a time gives lower performance than necessary. When writing to tape, the user can set fixed block size mode, e.g. by doing mt -f /dev/nst0 setblk 10240 Each tape I/O can then be a multiple of the block size, rather than a single block. On my system the Linux kernel imposes an arbitrary limit on I/O length of 512KB. But that would still allow each I/O to be of 51 10240-byte blocks. It would thus improve performance if star were able to, when writing to tape, write in chunks of 51 blocks (in this example). The final write would be for a lower number of blocks. The same could apply for reading from tape.