home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.unix.aux
- Path: sparky!uunet!destroyer!ncar!noao!noao.edu!tody
- From: tody@noao.edu (Doug Tody)
- Subject: Re: Yet more about 'tc' (blocking issues)
- Message-ID: <1992Nov11.032712.26391@noao.edu>
- Sender: news@noao.edu
- Nntp-Posting-Host: lepus.tuc.noao.edu
- Reply-To: tody@noao.edu
- Organization: National Optical Astronomy Observatories
- References: <1228@jagubox.gsfc.nasa.gov>
- Date: Wed, 11 Nov 1992 03:27:12 GMT
- Lines: 67
-
- In article <1228@jagubox.gsfc.nasa.gov>, jim@jagubox.gsfc.nasa.gov (Jim Jagielski) writes:
- > Is the fact that 'tc' needs to be blocked at 8k a big problem (or pain)
- > for most people? So far I've heard from only one person who says that it is,
- > but the only reason is because it makes it tough (impossible) to read tapes
- > from other workstations that write at 512/1024 and not because of any
- > A/UX-world problems.
-
- The 8k limitation is a minor nusiance for A/UX-only usage, e.g., file backups,
- but is a MAJOR problem if the tape drive is to be used to transfer data to or
- from an external, non-A/UX system.
-
- For example, at NOAO we have a data analysis package (IRAF) which can run on
- A/UX and is used to process astronomical image data. This data is
- voluminous, hundreds of megabytes, and is written to DAT or Exabyte tapes by
- the systems used to acquire the data, typically Suns. On these systems the
- standard way to write data transfer tapes is with a variable record size.
- For example we might write FITS tapes (an astronomy tape data format) using
- a block size of 28800 bytes per tape block. Given the large amount of data
- involved, tape is about the only practical way to transfer this data. Network
- transfer is also feasible, but is not an option for some users, and does not
- provide a backup capability.
-
- The blocking issue can also be a serious problem for other types of data
- interchange, for example tar tapes used to export software. We cannot
- currently write a DAT or Exabyte tar tape on any workstation class machine
- at NOAO (Sun, DECstation, VAXstation, IBM RS/6000, MIPS, SGI, HP, ...) which
- can be read on an A/UX system, due to the 8k blocking requirement.
-
- > I for one don't mind the 8k blocking at all. It makes the driver more
- > efficient (and therefore quicker) but you have to recall to block at
- > 8k. Of course, if 'tc' used the tape drive's actual physical block size
- > as it's blocking factor, then you'd have to block at either 512 or 1024
- > or 8192, depending on your tape drive so you'd still have to concern
- > yourself about blocking...
-
- A variable record size is preferred over a fixed block size since for some
- tape formats the record size may have some meaning, and since variable size
- records cause fewer transport problems. If fixed size blocks are used, one
- system may use, e.g., 1k blocks and the other 8k blocks, and it may not be
- possible to read the tape. By using variable size records it is possible to
- make an exact copy of a 9tk tape on a DAT or Exabyte.
-
- There is very little difference in terms of efficiency between fixed blocks
- and variable sized records for devices like DAT and Exabyte that use a large
- FIFO buffer internally to buffer the data. So long as this buffer is kept
- full during a write the tape will stream, writing data onto tape at the
- maximum rate. For cartridge drives like the DC2000 or QIC with no internal
- buffering a large block size is needed to transfer data efficiently, but
- this is better done in the application than in the driver. Rather than
- requiring 8k blocks, a good approach would be for the driver for a cartridge
- tape device to require that data transfers be a multiple of the physical
- device block size, e.g., 512 bytes, and leave it up to the application to
- buffer data to cause the tape to stream.
-
- Cartridge drives support only fixed size blocks, but DAT and Exabyte are
- capable of a variable record size mode as well as a fixed block mode
- (variable record size mode is enabled by setting the device block size to 1
- byte; in variable record mode each read returns one record). 9 track reel
- tape drives support only variable size records. For DAT and Exabyte a good
- driver will provide different device entries for fixed block mode i/o and
- variable size record i/o, so that the user can decide which to use. If the
- device is capable of compression, there should be different device entries
- for compression mode and non-compression mode as well.
-
- --
- Doug Tody National Optical Astronomy Observatories
- tody@noao.edu P.O. Box 26732, Tucson, Arizona, 85726
-