SUMMARY - Administration Workload

From: Edward Almasy (almasy@geomail.geomet.com)
Date: Tue Jan 02 1996 - 12:03:32 CST


My original query:

      What percentage of a full-time position is needed to administer a
   network of ten workstations (of mixed vintage and flavor: SunOS, Solaris,
   HPUX, Ultrix) and two PCs (running X and/or MS Windows) for an isolated
   group involved in scientific computing and commercial software development?
   5%? 50%? 500%?

The results:

   I received thirty responses, five of which offered useful commentary but
   no conclusions. Of the twenty-five responses that did offer a conclusion,
   the most common response was 100%, indicating that one full-time position
   would be needed to maintain the described network.

   Responses ranged from 5% to 200%, with a crude breakdown (rounded to 25%
   increments) as follows:

              0 x
             25 x x x
             50 x x
             75 x
            100 x x x x x x x x x x x x
            125 x x
            150 x x
            175
            200 x x

   Many people offered guidelines or advice on factors that may affect the
   administrative workload. Here are some of the most commonly mentioned:

       - user experience level (novice or expert users both greatly
           increase administrative workload)

       - amount of disk storage online (diskless workstations require
           less maintenance)

       - network heterogeneity (workload increases significantly as
           the number of OS or hardware platforms involved increases)

       - client/server versus peer-to-peer network model (decentralized
           disk storage increases required maintenance)

       - need for utility programs / scripts

       - number of applications installed (including "basic" packages
           like sendmail, perl, gcc, emacs, et cetera)

       - number of peripherals connected to the network
   
       - level of security required (electronic mail or an internet
           connection will substantially raise this requirement)

   Several people also noted that it is important for a system administrator
   to be proactive rather than reactive, and that this approach will require
   additional time (presumably to be returned as increased productivity for
   the group being served).

   I have included excerpts from a number of the replies below to expand on
   the above list and to offer further insight into the issues involved.

   I was also directed toward SAGE, the System Administrator's Guild, which
   maintains a useful web site at http://www.usenix.org/.

My thanks to the following people for their time and effort in responding
to this query:

        Andrei Ryzhov <Andrey.Ryzhov@Russia.Sun.COM>
        William C. DenBesten <DenBesten@cs.bgsu.edu>
        Henry Mensch <HMENSCH@us.oracle.com>
        Michael Bennett <Michael.Bennett@mail.esrin.esa.it>
        Michael Pavlov <Michael_Pavlov%CHASE%CHASE_NA_EXTERNAL@notes.worldcom.com>
        Andrew Lamb <andrew@mis.mua.go.th>
        Mike Barnett <barnett@zycad.com>
        Todd Boss <bfr7kq6@is000913.BELL-ATL.COM>
        Bob Corbett <bob@singapore.geco-prakla.slb.com>
        Dave Cain <cain@syrres.com>
        Christopher L. Barnard <cbarnard@cs.uchicago.edu>
        Celeste Stokely <celeste@stokely.com>
        Chris Eastman <chris@cwi.net>
        David Steiner <dsteiner@ispa.uni-osnabrueck.de>
        David Fetrow <fetrow@biostat.washington.edu>
        Guy Cole <guycole@netcom.com>
        Harel <hareldv@netvision.net.il>
        Bill Holzapfel <holzb@redwood.vkm.com>
        Josh Goldsmith <joshin@sei.com>
        Keith M Willenson <keith@oz.health.state.mn.us>
        Francois <leclerc@austin.asc.slb.com>
        Lyle E. Miller <lyle@ocs.com>
        Andrej Misik <misik@alpha.dcs.fmph.uniba.sk>
        Peter J. Welcher <pjw@ccci.com>
        Ric Anderson <ric@rtd.com>
        Rainer Leberle <rleberle@auspex.com>
        Stephen P Richardson <spr@myxa.com>
        Stephen Harris <sweh@mpn.com>
        Terry L. Perkins <tperkins@khis.Kodak.COM>
        Vahsen Rob <vahsenr@ce.philips.nl>

               -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-

From: celeste@stokely.com (Celeste Stokely)

In my experience, the # of workstations is only part of the equation--the
level of service expected by the users is an even bigger factor. One
mission-critical CAD development group who pounded the net and servers
of only 12 machines, 6 users, no deep unix knowledge rated their own full-time
sysadm. Kept me hopping, too!

               -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-

From: DenBesten@cs.bgsu.edu (William C. DenBesten)

Once you have more than a few workstations and have become dependant on
them, it is time for an administrator. If you don't hire one, one will be
carved out of your existing staff, an afternoon at a time.

If you're fighting for a systems administrator, I would suggest that an
algorithm similar to the following. The important point is that the
software is just as important as the hardware when calculating workload.
Units are percentage of an administrator (i.e. 5 min per day, 24 min per
week, or 2.5 days per year). All numbers inluded some slack for activities
such as keeping up with sun-managers.

10% to keep the first machine powered up
 5% if you care about security (e.g. are on the net)

 5% per application which is important, requires frequent updates, or
    is complicated to install (gcc, perl, sendmail, image magick, NFS,
    NIS, backups)
 1% per application which needs infrequent version updates,
    and configures really easily (e.g. gopher, tar, make, gzip)

 7% per workstation with user data on it.
 3% per dataless workstation which is not _exactly_ like one already
    accounted for.
 1% per dataless workstation which is _exactly_ like one already
    accounted for.

               -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-

From: "Henry Mensch" <HMENSCH@us.oracle.com>

how sophistocated are the users? if they're not very sophistocated, or if
they're very sophistocated in their uses, then they'll require more support
... in the former case, mostly handholding, and in the latter, mostly legwork
relating to rolling out new services as requested by each user.

               -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-

From: barnett@zycad.com (Mike Barnett)

I hear that Sun delegates one systems admin for 100 Suns and 10 to 12 servers.
But, that is a homogenous system.

I believe that the time required is logrythmic in equation; The more machines
added to a network, the more problems and fail point are introduced.

               -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-

From: leclerc@austin.asc.slb.com

. Each platform has got its own vendors's bug to track

. Need to be knowledgeable enough to know standard protocols (X, IP suite)
  to debug cross platform issues : time to learn/read

. "scientific computing and commercial software development"
  In my experience they are the most knowledgeable and challenging users,
  by the number of application they use (compilers, mathematical software,
  graphical output ... plus usual desktop application) and by their in-depth
  computer knowledge. Usually managers lightly using email and word processors
  can not apreciate that ...

. For your info, SAGE ( URL:http://www.usenix.org/ )
  (the System Administrator Guild) has got profile salary and job description
  information.

               -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-

From: Stephen P Richardson <spr@myxa.com>

        You are right, it is impossible to come up with a simple
        formula. Off the top of my head, I would have to consider
        things such as:

        How many end-users are being served?
        What is the level of experience of the end-users?
        Is the Sysadmin responsible to discover changes, or can the
            users be counted on to help?
        How often do the machine configurations need to be changed?
        Is the Sysadmin expected to install and test new releases
            as they are released?
        How much disk space is used? - this determines how often disk
            failures will need to be repaired and how tedious backup
            and tape management will be.
        Does the software/hardware configuration offer sufficient
            redundancy in function to allow repairs to be made on a
            non-emergency basis?
        Is the Sysadmin responsible to evaluate and keep users
            informed of the latest technologies?

        In general, my guess is that, with a cooperative group of
        end-users, one mid-level Sysadmin should be able to handle the
        task with occasional assistance from consultants or specialists
        for extraordinary tasks such as installing new products or
        major upgrades. In my experience, the most critical factor is
        the number and experience of end-users. One person can be
        consumed with simply training and correcting errors from even a
        small group of undisciplined users. The other major factor
        that I see in the environment you describe is the
        heterogeneity of hardware and software. This makes the
        Sysadmin's job *much* more difficult, and could easily double
        the resources needed to manage the systems.

               -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-

From: misik@alpha.dcs.fmph.uniba.sk (Andrej Misik)

        It depends on what type of administration you want: the approach
        "admin, we need to do this, please do it" I think will consume
        about (maybe less) than 50%. But if you need a person who will
        take care of your net in means of advancing it, providing
        better services and so on (I mean the approach "admin, here is
        the net, make it the best as you can" - which means administrator
        must keep eye on happenings in the Internet (installing new
        versions of software, act in response to new security vulnerabilities,
        thinking about the whole net and improving performance etc.))
        will IMHO need one full-time position.

               -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-

From: Ric Anderson <ric@rtd.com>

Once into configurations, one needs to spend time making access as
consistent as possible across the multiple architectures. E.G.,
common /etc/group (or NIS group) files, same uid for a given person
on each platform so that NFS file access will do the expected, etc.
Similar effort needs to be spent on logfile collection (are all
events being properly logged, and is someone reading those logs
daily?).

The person should next look into growth options so that they have
a concrete data allowing them to keep abrest of the needs of the
people trying to do "real work". This will mean some sort of
network analyzer (more capital) to allow monitoring of use levels
and establishing a baselines, so that when people start grumbling
about the network being "slow", one has solid knowledge as to "why".

The kind of person I would want in a job like this is not a typical
(from a personnel perspective) "network guru" or "OS wizard". The
individual needs to have enough hardware and software background
to determine what the underlying cause of a problem is so that it
an be dealt with in a timely manner. Too many "System Administrator"
and/or "Net Administrator" types lack this level of expertise; in
fact some pale visibly when asked to install or replace a memory
module or some other component.

Experience on multiple system types is a big plus (even if it isn't
on the systems you have) because it gives the person the perspective
needed to function properly in a mixed vendor environment. All
too often a person with several years of single OS experience (e.g.,
HPUX) will try to make everything (Ultrix, Sun,...) look like HPUX
because that's where he is comfortable. This is contraproductive,
but it'll be a year or two before he figures that out :-)

Further, the poor soul needs to be a decent installer. In small
companies one is often expected to extend the network to the storage
room that just became an office for someone. This will involve
running wire neatly and unobtrusively. If there is a false ceiling
or an attic this is simple. When surface runs are needed, some
knowledge of WireMold or other types of raceway products to produce
a neat (and protected) run is important.

               -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-

From: holzb@redwood.vkm.com (Bill Holzapfel)

 As you mentioned its hard to give an accurate estimate
because of the unknowns. But I can give you some guide
lines based on my experience:

1) How stable is the enviorment ? Both now and in the near
   future ? Are there many changes to either users or the
   software they run ?
   
2) Based ont the response to the above question how much
   research/testing needs to be done for the future ?
   
3) What kind or printers/plotters or other periphrials are
   there ? These tend to take the most time and the problems
   are usually of the critical nature.
   
4) What kind of applications do the users run ? Are they
   developed in house ? How much user support would this
   person be expected to give on the applications themselves ?

               -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-

From: Michael.Bennett@mail.esrin.esa.it (Michael Bennett SERCO ESA/ESRIN DPE/IO)

One person, full-time is quite capable of dealing with it. It's not so much
a matter of percentage, as it is of quality. Find a systems administrator
who knows at least 3 of the O/Ss and is 'CONFIDENT' in networking in such
a small amount of workstations. And this person, 100% of the time,
administering the workstations, should be quite capable of such an effort.

  ||| | | | | | | | | | | | | | | | | | | | | |||
   6510 Grand Teton Plaza Edward Almasy net: almasy@geomet.com
   Suite 310 Senior Software Engineer voice: 1 608 829 1242
   Madison WI USA 53719 Geometrics Corporation fax: 1 608 829 1243
  ||| | | | | | | | | | | | | | | | | | | | | |||



This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:10:50 CDT