Davina Treiber wrote:
Oh dear, so much misinformation in one thread. This subject has been
covered so many times before, yet people are still posting stuff that is
plain wrong. Let's try to set the record straight.
Jose Quinteiro <jquintei AT LEGATO DOT COM> from Legato, who should know better
and should have read the release notes and been on all the training posted:
Please forgive me. I thought looking at the source was sufficient.
Block size is set automagically by Networker based on device type. As
long as the devices are configured to be the same type, the block size
should be the same regardless of platform.
The environment variables are used to override Networker's defaults.
The defaults should be fine.
This is wrong. The default block sizes do vary between platforms for some
device types. This is because of the limitations of certain HBAs and their
drivers.
The only such limitation I am aware of is the one I mentioned on
Windows. Please enlighten me as to the other exceptions.
In any case, Networker is trying to set the same blocksize it would use
in other platforms, and the driver just can't handle it. It is, of
course, possible to mis-configure your OS (to use fixed-size blocks, for
example) to ignore what Networker wants. But Networker wants the same
device-specific block size, regardless of platform.
The same Legato employee then posted with apparent authority:
(snip)
FWIW, the "Not enough space" messages typically mean the tape is blank,
but you read a label sucessfully. Any errors in your system log?
He of course should have known that "not enough space" frequently means
that there is a block size mismatch. He also should know his employer's
stance on employees posting to this list with a Legato email address -
especially when what they are posting is garbage.
That's why I said "typically". That means "not every time."
Yura Pismerov then wrote:
Isn't it recommended to keep the default block size 0 (variable length)
in Linux ?
Yes it is, but this is because setting the block size to 0 allows NetWorker
to use the values configured. If your don't do this, Linux will only allow
you to used a fixed block size, and if this is only 512 bytes it will
seriously restrict the system's performance.
Linux's st driver does not default to fixed-size blocks for most tape
drives I've used. AIT-3 is a notable exception.
Kit Cunningham then suggested:
My experince is windows2k networker uses the variable setting for writing
only and the registry change for writing/reading. I do not know why.
This is wrong. The block size is set by setting the relevant environment
variable. However this won't work if the scatter/gather buffer settings are
insufficient, so it will just set it to a lower value - giving the effects
described. YOU NEED TO DO BOTH!
The block size is set internally by Networker. The environment
variables override the internal defaults. You absolutely DO NOT need to
mess with these environment variables! Please don't make things more
complicated than they need be.
and he continued:
I suggest ( if not already suggested by someone) you use st.conf entries
to set your blocksize.
This is a pointless blind alley, because st.conf doesn't set the block
size. The application (NetWorker) does this, provided that you have set
other factors in your operating system to allow this (e.g. scatter/gather
buffers, stinit.def and so on).
An improperly configured st.conf can cause all sorts of blocksize
troubles. If you don't set the ST_NO_RECSIZE_LIMIT flag the biggest
block Solaris will write will 64K - 2 bytes. Furthermore, it silently
breaks up bigger writes into these smaller blocks. So if you try to
write a 128K block you'll get three blocks; 64K - 2 bytes, 64K - 2
bytes, and 4 bytes. These tapes will have serious performance problems
recovering in NW 6.1.3 and later, and may be unrecoverable with some
combinations of Solaris/HBA in earlier versions of Networker.
The moral of the story is "don't mess with st.conf." Use Sun's st
patches. That's what Sun wants you to do, and they have good reasons.
I will send you my manager's contact information off-list. Please
direct any further complaints about my postings to the list to him.
Saludos,
Jose.
--
Note: To sign off this list, send a "signoff networker" command via email
to listserv AT listmail.temple DOT edu or visit the list's Web site at
http://listmail.temple.edu/archives/networker.html where you can
also view and post messages to the list.
=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=
| 