I am trying to figure out how disk space is used when overwriting backup
files.  Here is a description of our situation...

We have a 700+GB reporting database (SQL 2000) that we backup weekly to a 1
TB drive.  This has worked fine for quite a while but we have seen the
server crash twice in the last month while performing the backup.  The
backup is set to backup to 5 files which overwrite the previous weeks backup
(the old backup has already been backed up to tape at this point).   The
backups are run via a scheduled job, but when the server has crashed we see
no evidence that it ever ran in the job history.  The dba that set up the
job never set it up to go to a log file so we don't have that to look at
yet.  The only thing I do have are a great deal of messages in the SQL
Server log that occured while the last backup was running.   The following
messages began appearing a couple hours after the backup began and continued
until the server was rebooted.

SQL Server has encountered 3 occurrence(s) of IO requests taking longer than
15 seconds to complete on file [V:\MSSQL\Data\tempdev_2_Data.NDF] in
database [tempdb] (2).  The OS file handle is 0x00000598.  The offset of the
latest long IO is: 0x000004cab00000
Time out occurred while waiting for buffer latch type 4,bp 0x129ffc0, page
3:2305080), stat 0x40d, object ID 2:1682105033:0, EC 0x2015A3F0 : 0,
waittime 300. Not continuing to wait.
Waiting for type 0x4, current count 0x100022, current owning EC 0x2015A3F0.

Just some additional info on what I mean by "the server crashed".  This
server is both a database server and a webserver which runs and admin
website used to monitor database extracts.  At some point the server stopped
responding to all forms of remote control, and the actual screen in our
datacenter went black.  No one could log onto the server or see anything.
Enterprise manager seemd to be connecting on, although it was extremely
slow, and the website was working as well.  When the 4 hour backup had run a
full 12 hours over the normal time we had the datacenter do a hard power
down and restart the server.  Everything came up fine.

Getting back to the backups, and my question...

I know when I run backups on large databases I will see the zero byte
file(s) created and the size doesn't appear to change until the backup is
complete.  I'm wondering if the data is somehow being cached or stored in
temporary files until all the data is available then it written all at once.
I can't imagine SQL Server is caching 750GB, and if it uses temporary files
on the disk I would think my backups should have been failing for months.
It was suggested by our server admins that it may be overwriting the
previous backups but not updating the file system tables to show this, which
means we would have corrupt backup files on the server.

For the time being I have the job set up to delete the previous backups
prior to beginning the new backup, which I hope alleviates the problem.
Regardless of whether this works or not, I want to understand what is
happening and what my disk space requirements are going forward.

Can anyone explain how the disk space would be utilized in this situation?

Thanks in advance.
- Jim