[WEB4LIB] Purchasing a server [long!]

[James Klock]

>Here's my question: If I know I only want one processor and a single hard
>drive on my machine (we're a very small library), is there any reason to go
>with a server rather than a PC with an internal tape back up and UPS?

Generally, there is a significant difference in quality between a server
sold by Dell/IBM/Gateway/Compaq/Micron/... and a PC sold by the same
manufacturers, let alone a PC sold by Joe's Computers on the corner of your
local town square.  When you're buying a PC that will sit on someone's desk
and surf the web, this difference in quality is generally irrelevant.  When
you're going to ask a machine to do lots of different things for lots of
different people all at the same time (provide this word document to Fred
while Jane prints out this other Excel spreadsheet to the network laser
printer and Ricky searches through his home directory for that other file
he wanted), how fast it is able to access and retrieve information from
disk and from memory can make a difference, even when there are a small
number of users.  Basically, that means SCSI disks, good memory and a
motherboard that will provide fast I/O to the processor (look for the words
"100MHz bus").  

Your needs also depend, somewhat, on how important it is that the server be
working at any particular moment....  When Dell sells you a PC, they
generally use high-quality parts.  When Joe's Computers sells you a PC,
they may or might not.  When 10 or 15 people are all depending on the
server on a daily basis, it becomes unacceptable for that server to be
unavailable, even for a little while, with any regularity.  So you want to
get good parts, especially for the SCSI controllers (there are quite a few
inferior controllers out there.  Adaptec and BusLogic both are generally
highly recommended).  You might also want to consider some common "fault
tolerance" schemes, which will allow you to keep on using the server even
when something breaks, so that you can go in and fix it at a more
convenient time.  Examples include using two identical hard drives and
"mirroring" them (this is supported at the software level by most Network
Operating Systems, such as NT Server and Novell Netware, or can be done by
some SCSI controllers as a "software-controlled RAID-1 array").  Mirroring
is particularly effective if you have the disks on different controllers--
that way, if one controller fails, the other will go on talking happily.
Other fault tolerance schemes include using separate disks for the OS and
the user data.  This allows you to distinguish easily between the stuff
that doesn't change very often and is easily reproduced (the OS disk, which
only changes when you reconfigure the system, and can theoretically be
rebuilt from scratch from the installation media) and information that
cannot be easily reproduced and is more critical (the user data).  
Given the cost of disks, adding an extra hard drive or three shouldn't be
too expensive, even if they're high quality SCSI disks, and might save you
a LOT of trouble in the future.  

Sorry that got so wordy!

James