SUMMARY: Slow TAR from SUN

From: Lauriault Yvon <yvon.lauriault_at_nlc-bnc.ca>
Date: Fri, 14 Mar 2003 14:48:08 -0500

Thanks to all who responded.

Turns out the blocksize used to create the tape was 512 bytes. It was
suggested to increase it to 64K to improve the throughput.

Yvon

----------------------------------------------------------------------------

--
From: Haesaerts, Corinne [Corinne.Haesaerts_at_hp.com]
better ask what blocksize was used. Was it just TAR or did it involve DD ??
----------------------------------------------------------------------------
--
From: Rick Kelly [rmk_at_toad.rmkhome.com]
Whoever created that tape used a block size of 1. The normal default
block size is 16k. Unfortunately you can't change it.
----------------------------------------------------------------------------
--
From: Alan Rollow - Dr. File System's Home for Wayward Inodes.
[alan_at_desdra.cxo.cpqcorp.net]
	It is most likely the small block size (512 bytes per
	tape block).  You can check by getting tar out of the
	way and reading the tape directly with dd(1):
		time dd if=tape-device-name of=/dev/null bs=1b
	If you use the no-rewind tape device, this will avoid
	timing the rewind.  For a small archive this might add
	noticable time.  If you want to limit the amount of
	data read to get your performance estimate, use the
	"count=" option on the dd(1) command.
	I think the only way to get this to read faster is to
	have the tape rewritten with a useful block size.  tar
	defaults to 10 KB on most systems and that's not bad
	for a DLTtape III.
	With the a DLT III tape, the performance will limited
	to around 1.5 MB/sec, though it may peak higher with
	compression (PDF and TIFF may already be compressed so
	that won't help).  DLT IV can go higher with a decently
	written tape and the right data stream.  With the DLT IV
	tapes, the TZ89 like large block sizes.  256 KB isn't bad
	if you can coerce tar to write it.
----------------------------------------------------------------------------
--
From: Dr Thomas.Blinn_at_HP.com [tpb_at_doctor.zk3.dec.com]
If the tape was truly created with a blocksize (blocking factor)
of 1, then it has 512 byte records, and nothing you do on your
system is going to speed up the data rate for reading it.  You
need to get the tape written with a higher blocking factor; at
least 10, maybe more like 50, or even 64 if the Sun system can
support 8K data writes to tape.   
----------------------------------------------------------------------------
--
From: Alan Rollow - Dr. File System's Home for Wayward Inodes.
[alan_at_desdra.cxo.cpqcorp.net]
	Just as an extra bit of information...
	I used dd(1) to write a copy of /vmunix (9,141,408 bytes) to
	my TZ87.  With a DLT III the performance will be comparible to
	the TZ89 using the same kind of tape.
	Writing at   1 sectors per tape block took 91 seconds: 98 KB/sec
	Writing at 128 sectors per tape block took  7 seconds: 1275 KB/sec
	Reading of each:
		512 bytes  ->  122 KB/sec
		 64 Kbytes -> 2231 KB/sec
	The performance you saw is to be expected.  Take comfort in
	the fact that it probably took the other side longer to write
	it.
----------------------------------------------------------------------------
--
-----Original Message-----
I have received a DLT III tape (density 10/15) that was created using TAR on
a SUN server and find that it is very slow to read (120 KB/sec) using a TZ89
(35/70) drive connected to a GS140 Tru64 v5.1 pk5.
 
The data on that tape is TIFF images imbedded in PDF files with an average
of 4MB each. I noticed that when I started TAR it displayed "blocksize = 1".
 
Is 120 KB/sec normal with this setup?
If not, any suggestions to improve the throughput?
Blocksize maybe? What value?
 
 
Thanks in advance.
 
yvon.lauriault_at_nlc-bnc.ca <mailto:yvon.lauriault_at_nlc-bnc.ca> 

Received on Fri Mar 14 2003 - 19:47:00 NZDT