how does the scsi chain work?

[Kenneth D. Merry]

On Fri, Mar 11, 2005 at 17:16:59 -0500, Dan Langille wrote:
> Hi folks,
> 
> I added the following devices to my computer, all in one SCSI chain:
> 
> sa0: <QUANTUM DLT7000 2561> Removable Sequential Access SCSI-2 device
> sa0: 20.000MB/s transfers (10.000MHz, offset 15, 16bit)
> sa1 at sym0 bus 0 target 6 lun 0
> sa1: <QUANTUM DLT7000 2255> Removable Sequential Access SCSI-2 device
> sa1: 20.000MB/s transfers (10.000MHz, offset 15, 16bit)
> sa2 at sym0 bus 0 target 8 lun 0
> sa2: <QUANTUM DLT8000 0114> Removable Sequential Access SCSI-2 device
> sa2: 20.000MB/s transfers (10.000MHz, offset 15, 16bit)
> 
> Then I decided to do some concurrent testing.  See below for details. 
>  The goal was really to test filling the tape. This exercises both 
> the drive and the tape (all of which are used).
> 
> Sometimes a single tar takes nearly 4 hours, sometimes it takes only 
> 75 minutes.
> 
> My question:  It doesn't look like the load is shared evenly.  Can 
> someone talk a bit about this?  That's a pretty general and wide 
> question, I know...

You might want to look at your media quality.  You said that your drives
and media are used...

In this case you even got a write error.  That's pretty indicative of bad
media.  (Or a drive that needs to be serviced I suppose.)

> [root at lux:/home/dan] # date && time tar cf /dev/nsa0 /usr  && time 
> tar cf /dev/nsa0 /usr  && time tar cf /dev/nsa0 /usr  && time tar c
> f /dev/nsa0 /usr  && time tar cf /dev/nsa0 /usr  && time tar cf 
> /dev/nsa0 /usr ; date
> Thu Mar 10 22:31:57 EST 2005
> tar: Removing leading '/' from member names
> 
> real    75m15.394s
> user    0m45.666s
> sys     2m31.039s
> tar: Removing leading '/' from member names
> 
> real    114m5.702s
> user    0m52.549s
> sys     2m45.939s
> tar: Removing leading '/' from member names
> 
> real    156m18.091s
> user    0m51.538s
> sys     2m46.858s
> tar: Removing leading '/' from member names
> 
> real    119m48.240s
> user    0m49.670s
> sys     2m45.721s
> tar: Removing leading '/' from member names
> tar: usr/home/bacula-restores/usr/X11R6/share/locale/bs/LC_MESSAGES: 
> Write error
> 
> real    195m0.408s
> user    0m38.779s
> sys     2m13.158s
> Fri Mar 11 09:32:26 EST 2005


And you got another write error on this tape:

> root at lux:/home/dan] # date && time tar cf /dev/nsa1 /usr  && time tar 
> cf /dev/nsa1 /usr  && time tar cf /dev/nsa1 /usr  && time tar cf 
> /dev/n
> sa1 /usr  && time tar cf /dev/nsa1 /usr  && time tar cf /dev/nsa1 
> /usr ; date
> Thu Mar 10 22:53:40 EST 2005
> tar: Removing leading '/' from member names
> 
> real    103m35.344s
> user    0m50.394s
> sys     2m44.732s
> tar: Removing leading '/' from member names
> 
> real    112m40.647s
> user    0m50.555s
> sys     2m46.049s
> tar: Removing leading '/' from member names
> 
> real    152m45.638s
> user    0m52.656s
> sys     2m46.040s
> tar: Removing leading '/' from member names
> 
> real    201m0.014s
> user    0m53.838s
> sys     2m51.051s
> tar: Removing leading '/' from member names
> tar: usr/home/bacula-
> restores/usr/src/contrib/ncurses/form/fld_link.c: Write error
> 
> real    93m59.622s
> user    0m22.483s
> sys     1m8.356s
> Fri Mar 11 09:57:41 EST 2005

With a tape, if it can't write in one area, it'll skip down the tape to try
to find an area it can write to.  So you may just be running into bad areas
on the tape.

I would recommend buying some new media that is speced to work on those
drives, and then try it out.  If you still have variable performance and
write errors, then you may have a drive problem.

Don't forget to run a cleaning tape through if those drives use cleaning
tapes.

Ken
-- 
Kenneth Merry
ken at kdm.org