News from PANUG - August 21, 2001
http://panug.org

LONG TERM DATA STORAGE
by Ed Sawicki

I've always known that magnetic media, such as diskettes
and tape, can only store data reliably for so long before
the media becomes unreadable. I've just not known how
long. Recently, I discovered the answer.

A few months ago I started doing backups to optical media
- inexpensive CDRs. At about 35 cents each, I do my
weekly backups to CDR and daily backups to another hard
disk on the network. I started going through my diskette
and tape archives to see if there was anything I wanted
to move over to optical. There were data on diskettes and
tapes that were not terribly important (memorabilia) yet
I wanted to preserve.

About one-third of the diskettes and tapes had problems -
mostly sector not found errors when I tried to read them.
They were originally recorded in 1985 and most had never
seen the inside of a drive since then. I recovered as much
data as I could and burned it all on a CDR.

Since I still use diskettes, I tried to format them so I
could reuse them. I figured that the data I recorded on
the drive 16 years earlier had "eroded" away but the
media was still good. Not so. About one-half of these
diskettes couldn't be formatted with Windows 2000 though
I tried each diskette three times. I then used Linux to
format the diskettes that W2K wouldn't and about half could
be formatted. For some odd reason, Linux is better at
formatting diskettes than W2K - and does it quicker.

The message here is that not only does data erode over
time, the ability of the media to store newly-written
data diminishes. I experimented with diskettes and tapes
that were not as old and I estimate that you can safely
store data for about 10 years. After that you better have
two or more copies of the data. Better yet, switch to
optical media that is supposed to store data for 50 years
or more.


VIRTUAL SERVERS

One of the subjects that will be covered in the Apache
Boot Camp on September 5-6 is virtual servers. This allows
you to cluster two or more servers (web, email, etc.)
together such that:

a) requests are evenly distributed amongst the servers in
   the cluster.

or

b) all requests go to the main server until it fails; then
   requests are automatically sent to the backup server(s).
   The switchover time is nearly instantaneous.

While virtual server clusters are commonly used with Apache,
the technique can be used with any servers that have IP addresses.
For example, you can do this with a cluster of computers running
Microsoft's IIS without having to buy in to expensive vendor-proprietary
clustering products. This form of clustering is easily achieved with
several open source software packages. In one case, the software
was installed and setup in less than 5 minutes.

For more information on the Apache Boot Camp, visit the PANUG
web site at http://panug.org