Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!ucbvax!ICAEN.UIOWA.EDU!dbfunk
From: dbfunk@ICAEN.UIOWA.EDU (David B. Funk)
Newsgroups: comp.sys.apollo
Subject: Re: TAR (SR9.X), cartridges
Message-ID: <8905090651.AA01180@icaen.uiowa.edu>
Date: 9 May 89 06:51:14 GMT
Sender: daemon@ucbvax.BERKELEY.EDU
Organization: Iowa Computer Aided Engineering Network, University of Iowa
Lines: 67

In article <431a057f.10b48@apollo.COM> weber_w%apollo.uucp@eddie.mit.edu
  (Walt Weber) writes:
>In article <503@zgdvda.UUCP> pfuetzner@zgdvda.UUCP (Matthias Pfuetzner) writes:
>>I've got a question concerning TAR under SR9.2 and SR9.7:
>
>>3.) Is there any possibility to SPEEDUP cartridge tape actions?
>>Writing about 8 Mbyte to cartridge tape on a standard Sun 3/60 takes 
>>about 8 minutes (/dev/rst8), reading them bak with tar on an APOLLO
>>(DN 3000, DN 580) takes about 90 minutes (no jobs on the machine, no 
>>jobs on the network!!)
>
>Conventional wisdom (a.k.a. "folklore") is that writing to disk will always
>take longer than reading, since new files need to have space "found" for
>them, rather than chasing pointers to existing blocks as in the read operation.
>An increase of slightly over 11x "feels wrong", and since we (as users) always
>want it faster, less expensive, and available now, can you tell me how long
>it takes to WRITE that same 8Mb to disk (under a new pathname) on your 3/60?
>And how long does it take us to write those files from the DN3000 to ctape?
>( No sense fixing one tire if we've got two flats :-)


Pre-sr10 streams I/O to/from cartridge tapes is one big flat tire! ;-)
Utilities like /com/wbak, /com/rbak, & /systest/ssr_util/cptape do not
use streams I/O to get to ctape, they use an unreleased interface, the "ct_$"
calls. Their performance is OK, about 1 Megabyte/minute. Utilities like
/com/rwmt & tar use the streams I/O system, and their performance is
hurting, about 0.1 Megabytes/minute.
For example:

# DN300 node, 4 Mbytes memory, 150 Mbyte disk
#
#
# AEGIS8-DOMAIN/IX kernel, revision 9.7.0.4 , Tuesday, May 31, 1988   4:51:23 pm.
# $ ld -a file
#
# sys   type      blocks  current
# type  uid         used   length   attr rights       name
#
# file  obj         4791   4884620  P    pgndwrx      file
#
# $ rwmt -w -dev ct -f 1 -unlab -raw -rf f -rl 512 -bf 1 file
#
# Run time to transfer from disk to tape with streams I/O: 25.2 minutes.
#
# $ cptape -dev ct -read tape.cpy
#
# Run time to transfer from tape to disk with "ct_$" I/O: 3.8 minutes

At sr10.1 there is a big improvement in streams I/O speed for ctapes.
It's now almost as fast as the "ct_$" system. Taking the same node that
was used in the previous case, running it under sr10.1, and transfering
the same file, we get the following results:

# Domain/OS kernel(8), revision 10.1.0.2, January 4, 1989  10:48:48 am
#
# $ rwmt -w -dev ct -f 1 -unlab -raw -rf f -rl 512 -bf 1 file
#
# Run time to transfer from disk to tape with streams I/O: 5.5 minutes.
#
# $ cptape -dev ct -read tape.cpy.2
#
# Run time to transfer from tape to disk with "ct_$" I/O: 3.6 minutes

So the bottom line is, if you want to use tar with cartridge tapes
use sr10 or be prepared to do a lot of waiting.

Dave Funk