Exabyte question

Chris Kranz clk at splash.princeton.edu
Sun Dec 9 09:07:00 AEST 1990


I have an EXAbyte 8200 external drive (SCSI) on a 4/490 (SunOS 4.0.3) and
I have a couple of questions.

Does anybody know what parameters to give to dd to get the Exabyte drive
to act like a streaming device?  I'm trying to maximize the data transfer
rate and minimize the time it takes to write/read a file.

I've done some experiments with an 8 MB file and the data is not easy to
interpret.  Here are some examples:

        time dd if=ps.a.tar of=/dev/nrst1 bs=10k
        783+1 records in
        783+1 records out
        0.0u 3.4s 1:23 4% 0+208k 983+1io 980pf+0w

        time dd if=ps.a.tar of=/dev/nrst1 bs=20k
        391+1 records in
        391+1 records out
        0.0u 1.8s 1:24 2% 0+208k 2+1io 0pf+0w

        time dd if=ps.a.tar of=/dev/nrst1 bs=100k
        78+1 records in
        78+1 records out
        0.0u 1.0s 1:21 1% 0+296k 2+0io 0pf+0w

        time dd if=ps.a.tar of=/dev/nrst1 bs=246k
        31+1 records in
        31+1 records out
        0.0u 1.1s 1:33 1% 0+416k 2+1io 0pf+0w

The documentation gives the following information:

        - physical block size is 1K
        - data buffer consists of 256 KBytes of DRAM
        - maximum burst data transfer rate does not exceed 1.5 MBytes/sec
        - typical performance is expected to be 1.2 MBytes/sec
        - sustained transfer rate of 246 KBytes/sec

Why the difference between the "typical performance" rate and the
"sustained transfer" rate?  Which should I expect to see?  How can I
measure this? If I use the data above from time, which time do I use, user
cpu seconds or elapsed time?  Niether seemed to give the documented
transfer rate.  What am I doing wrong?

If anyone could shed some light on this mystery I would be very grateful.
Thanks.
            Christopher L. Kranz                     GFDL
  Internet Address: clk at gfdl.princeton.edu        Commercial: (609)452-6585
      UUCP Address: princeton!gfdl!clk                   FTS:      298-6585



More information about the Comp.sys.sun mailing list