August 1993 “Preliminary Hardware Configuration for a Main Service Linux Machine”

From 1992 through to 1994 I was working at the Computer Unit at the University of Wales (well, wrangled an “Employment Training” position there on my own initiative) as a sysadmin and was running Linux on an IBM XT (from very dodgy memory). A friend of mine, Piercarlo Grandi, suggested to me (semi-seriously I suspect) that you could now build a large enough PC to support quite a number of users, and that the Computer Unit could use it as a central server (they were running DEC 5830s with Utrix), so I knocked up a text file and discussed it with my colleagues. They didn’t take it very seriously – little did any of suspect how much that would change.

Well tonight I indulged in a bit of computer archaeology and managed to get the data off my Amiga hard disk (from a GVP A530 expansion unit) and browsing around happpened to stumbleover that text file, dated 8:20pm on the 8th August 1993. It’s quite touchingly naive in places, and my numbers are pretty ropey.. 🙂

Preliminary Hardware Configuration for a Main Service Linux Machine

Item                            Each    Number          Total

Case                             100    1                 100
Keyboard + Mouse                 100    1                 100
Floppies                         100    1                 100
DAT Drives                       750    4                3000
EISA SCSI Controllers            300    2                 600
Memory (Mb)                       25    256              6400
Pentium EISA Motherboard        1000    1                1000
3.5Gb SCSI-II Disks             1800    5                7000
Screen+SVGA Card                1000    1                1000
EISA Ethernet Card               200    2                 400
CD-ROM Drive                     300    1                 300

                                                        20000

Projected to be able to support between 200-400 users running Linux 0.99.p12
        (Alpha release kernel with patched IP - appears stable)


Notes:

(1) I've seen reports that the ethernet driver code may suffer from a
    memory leak, but I've not seen any evidence for this yet as my
    machine hasn't been turned on for a long enough period for it to
    cause any problems.

(2) As it is so new there is very little commercial software available
    for it, but there is a quite sizeable free software base with many
    of the GNU packages already ported for it, and this is generally of
    high quality.

(3) The Linux kernel is well thought out, and includes support for shared
    libraries (which Ultrix sadly never picked up) which significantly
    reduces the amount of memory applications need.

(4) A Linux box of the size proposed for the service machine has not
    been attempted yet (as far as I know), but ones of the size of the
    proposed testbed machine are already in usage on the Internet. I
    believe that Linux can handle this scaling up with no problem.

(5) There are apparently companies within the UK who sell support services
    for Linux, I will investigate further.

(6) There is already a large amount of Linux expertise on the Internet,
    including the comp.os.linux newsgroup, the linux-activists mailing
    list and even an IRC channel dedicated to Linux users.

This post is dedicated to Rob Ash, my then boss, who took a chance taking me on after my time as a student mucking around on computers when I was meant to be doing my Physics degree, and who was a great mentor for me.

One thought on “August 1993 “Preliminary Hardware Configuration for a Main Service Linux Machine”

  1. For the memory, at least, that’s a halving of cost every two years; fairly close to Moore’s law trend.
    No, now I come to think of it, I’m comparing disc memory to flash memory — disc-to-disc would be 16 halvings in 20 years, or 1.25 years. Kryder’s Law?

Comments are closed.