Robin Miller uhs wrote:
>
> Hi All,
> I think Tony has already expressed a good reason for larger transfers.
> If only for compatability with other operating systems so you can read their
> tapes, this in itself is a good enough reason.
Maybe. Maybe not. If the only tape drive under NT or other unixes that uses
a greater than 1MB block size is their tape drive, then it isn't to be
compatible with NT or Solaris, it's to allow the their hardware to work on
linux as well as the other OSes. It would not be any general compatibility
issue at that point, but an item of specific support for their hardware
instead.
> But even it this tape drive
> requires large transfers for "best" performance, then Tony should be able to
> tune things to his liking.
He can tune away on his personal copy of linux, but when it comes to the
aic7xxx or st drivers in the main stream kernel, then the change has to meet
the standards of peer review and be accepted for its merits. In this case,
that means he has to be able to justify why he would want to make a change
that would drastically increase the amount of non-pageable kernel memory that
the various drivers should allocate in order to satisfy the command
requirements of his tape drive. That justification then is subject to peer
approval. None of us are obligated to accept his tuning.
> On Tru64 Clusters, we use Legato Networker for backups. It normally
> uses multiple I/O's each of 256KB for best performance. Tru64 Unix also allows
> maximum transfers to whatever the tape drive supports which can be upwards of
> 16MB-1. The max transfer size, and many other attributes, are all tuneable
> via a device detabase.
>
> I personally don't think Tony's request is out of line.
>
> My 2 cents worth,
> Robin
You know, unless the tape stops writing and then has to start back up again
(the phenomena commonly known as "stops streaming") then regardless of the
tape block size, you should get the same overall throughput to the tape (tapes
don't vary their stream speed, so as long as it never stops streaming you are
getting full media throughput). So, what I want to know is if Legato
Networker is just using 256KB I/O ops to smaller tape block sizes (aka, more
than one tape block per I/O) or are they actually modifying the tape block
size on the tape device? If they aren't changing the tape block size on the
tape device, but are just sending fewer commands of larger size, then your
performance improvement would have to be from reduced command overhead in the
SCSI stack, not from improved tape streaming performance. That would be an
entirely different matter than how things have been presented so far.
--
Doug Ledford <[EMAIL PROTECTED]>
Opinions expressed are my own, but
they should be everybody's.
-
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to [EMAIL PROTECTED]
