Amanda-Users

Re: RAIT in 2.4.3b4

2003-01-29 22:30:11
Subject: Re: RAIT in 2.4.3b4
From: "John R. Jackson" <jrj AT purdue DOT edu>
To: Scott Mcdermott <smcdermott AT questra DOT com>
Date: Wed, 29 Jan 2003 21:56:17 -0500
As I was writing the reply to how to mirror to disk and tape, I realized
doing the async I/O is going to be a bit harder to integrate than you
might think.

The RAIT code does not call the OS write() function.  Instead, it calls
the tapeio library tapefd_write() call.  The reason is that some of
the RAIT segments may not be going to tape at all.  So tapefd_write()
may end up calling tape_tapefd_write(), or file_tapefd_write(), or
null_tapefd_write(), or ...

To do this right, we'd probably need to add tapefd_aiowrite() to the
tapeio library and the matching XXXX_tapefd_write() calls to the various
drivers (and the corresponding *aioread() and *aiowait() routines).
That's not a terribly big deal, but it's not the "trivial" case of just
adding a few lines in output-rait.c that one might think.

John R. Jackson, Technical Software Specialist, jrj AT purdue DOT edu

<Prev in Thread] Current Thread [Next in Thread>