🚀 go-pugleaf

Programme for UKUUG System Admin and Mgmt meeting
=================================================



Here is the running order for the UKUUG System management and
Administration Meeting to be held at Imperial College, London, 9th September
1993. The detailed abstracted are given after the summary listing.

The cost of attendence, including lunch is 47 GBP, inc VAT, for members
of UKUUG, SUKUG or EurOpen.

Bookings may be made by contacting the UKUUG Secretariat, 0763 273475
(voice), 0763 273255 (fax). Credit cards are accepted.


In the case of papers with multiple authors only the speaker is given.



Summary timings
===============

09:00	Registration and Coffee

09:25	Welcome and Introduction

09:30	Invited speaker: Evi Nemeth (University of Colorado, USA)

        SA-Tool: A System Administrator's Cockpit

10:30	BREAK

10:45	Stuart McRobert and Lee McLoughlin 

        The Archive

11:45	Piete Brooks

        Setting up the Network Time Protocol and Software updates
        using (e.g.) rdist

12:30	Clive King

        Filesystem change - a strategy to maintain user satisfaction!

13:00	LUNCH

14:00	Lindsay Marshall and Peg Schafer (SAGE/US)

        SAGE/UK 

15:20	Graham Dean

        The Devolution of Management

15:50	BREAK

16:05	David Lomas 

        What about the Workers!

16:35	Adrian Pell

        So I've got the data ... what now?

17:05	Robert Evans 

        User Management in the Imperial College Distributed Computing
        Environment

17:35	FINISH


17:45	UKUUG E.G.M.


Detailed Programme and abstracts
================================

Some papers have multiple authors, only the speaker is given below.

09:00	Registration and Coffee

09:25	Welcome and Introduction, Neil Todd (GID Ltd)

09:30	Invited speaker: Evi Nemeth (University of Colorado, USA)

Evi Nemeth, a faculty member in Computer Science at the University of
Colorado, has managed Unix systems for the past 15 years. She is
co-author of the best-selling "Unix System Administration Handbook"
(Prentice-Hall, 1989) and has taught numerous system administration
tutorials. Other interests include network measurment tools and
undergraduate education. She is secretary of the Board of Directors of
the USENIX Association.


SA-Tool: A System Administrator's Cockpit
+++++++++++++++++++++++++++++++++++++++++

Monitoring large numbers of machines in a distributed environment can
be time consuming and inefficient. Often a system administrator finds
there is a problem with one of his machines or network links by a phone
call from a frustrated user. satool provides a way to efficiently
monitor groups of machines and find problems and potential problems
quickly; it's sort of an early warning system for sysadmins.


satool is composed of three independent parts: an SNMP (Simple Network
Management Protocol) agent that runs on each machine to be monitored, a
database that collects data from each client machine, and a graphical
user interface (GUI) that acts as the interface between the user and
the data base.

In designing satool we have tried to adhere to two fundamental
principles: remain totally configurable (outguessing a sysadmin's needs
is hopeless) and reuse existing tools where possible. On the data
gathering side, configurable items include: the actual data to be
collected, how often to collect it, threshold values that indicate a
problem, the action to take if a threshold is exceeded. Configuration
for the GUI selects the hosts or devices to display, the type of widget
to represent a variable, and personal preferences for items like the
arrival of an alarm condition (for example color, sound, blinking,
etc). We use the tcl and tk languages from John Ousterhout at the
University of California, Berkeley for the display. An SNMP agent and
sysadmin MIB (Management Information Base) provide for communication
between the data collection and data gathering activities.

The actual data gathering is done by an SNMP agent running on the hosts
being monitored. We have extended the CMU SNMP 1.1b agent and MIB to
include variables of interest to UNIX system administrators not
included in the standard network or host MIBs. SNMP agents update a
variable when its value is requested. Thus, instead of the agent
sending information to the server at regular time intervals, the server
polls the agent when it wants a new value of a variable. (The one
exception is when the agent first starts up. It sends a trap to the
server to alert it that the agent is alive and can now be monitored).
Since the responsibility to request data falls on the server, the
server has complete control over what data it requests from the agent
and how often it requests it.

The satool server maintains a database of client machines and the
values of supported variables gathered during the last poll. Normally,
a client is added to the database when the server receives a message
from the SNMP trap daemon that the client's SNMP agent is running.
However, the server also maintains an on-disk copy of the database that
is read in on invocation of the server. Because of this, the database
will survive system crashes. Client machines can also be specified in
the server's configuration file to "prime the cache" of machines to
monitor. The time between polls is configurable on a per-machine basis
(overriding a stated default). The server listens for database requests
from the GUI on port 0xFB1.

The satool display system includes an X based GUI (Graphic Users
Interface) built using tcl/tk. It supports a hierarchical top level
view of the machines being monitored. Machines can be grouped to allow
the screen area to scale with the number of machines being watched. For
example, a sysadmin can configure his workstation to display the
machines he is directly responsible for, the machines that his
colleague on vacation is responsible for, the printer down the hall
that he uses regularly, and his network gate way to the outside world.
After traversing the hierarchy to an individual machine, the display
contains visual widgets representing the health of that machine: disk
space free, memory statistics, cpu activity, mail queue length, nfs
statistics, etc. The user can choose the form of the widget (currently
a number, a thermometer display, or a sliding window histogram) to
display each quantity.

Alarm conditions for each variable can be expressed in the
configuration files as well. If the value of a variable exceeds the
alarm threshold, a special action (blinking, beeping, reverse video,
digital pager, etc.) is taken on the screen to indicate it. Alarms are
propagated to the toplevel display, thus if /var/tmp fills up on host
heineken on the beer subnet of the cs domain and if the groups are set
appropriately, the alarm condition would be noted in the widgets
representing heineken, beers and the cs domain. A user would see or
hear the alarm independent of which level he was actively displaying at
the time.

A working prototype of satool exists. It will be used this fall by the
Computer Science Department's undergrad lab sysadmin group and
graduate/faculty research sysadmin group to monitor about 300 machines.
We expect to use feedback from these groups to expand the sysadmin MIB
and to build more display widgets. The code will be freely available
with a Berkeley style copyright notice.

10:30	BREAK

10:45	Stuart McRobert and Lee McLoughlin 

	(Dept Computer Science, Imperial College)	

This section covers The Archive - what it is, how to access it, what
services are available, and up to date news all about it.

The Archive
***********

The Archive contains publicly available information and software for
Unix Systems, PCs and Apple Macs, along with Technical documents e.g.
RFCs and lots more amounting to around 15 GBytes of generally
compressed data used extensively by some 15,000 hosts around the world
and ships by ftp over a GByte a day.


A tour of The Archive will be given showing interesting some places!
along with an update about the archive, where we are, where we are
going, and new interesting software interfaces e.g. WWW, WAIS and maybe
Gopher.

Software Distribution Service
*****************************

Announcing the new CD service - what is available and how to order
fresh CDs. Status of the older tape based service. How to request
special CDs, costs etc.

Archie 3
********

News about the updated Archie 3 server, problem fixes and future plans,
access methods.

11:45	Piete Brooks (Computer Laboratory, Cambridge University)

Setting up the Network Time Protocol
************************************

NTP allows a number of machines connected by an IP network to
synchonise their clocks to sub- millisecond accuracy. It's generally
useful to avoid having clocks out by minutes, but it's also annoying
when "make" fails due to clock diferences. There are an increasing
number of protocols (e.g. Kerberos and X.500) which use timestamps, and
thus need a reasonable time base. When tracing calls/logins between
busy machines, accurate time stamps make things much easier.

This talk goes through some of the basics of how it works, and then
describes how to set up a site for NTP, using the JIPS NTP Pilot as an
example.

Software updates using (e.g.) rdist
***********************************

This is not a talk with a high tecnical content, nor does it contain
any revolutionary ideas, it's just a `this is what we have done' talk
to stimulate others to exchange ideas.

We maintain a directory of trees for various "classes" of machines,
containing files that are not part of the "standard" distribution for
the relevant class of machine. From this is built "rdist" files, and we
then use a perl script to do a controlled parrelel rdist to the
clients. This can be adapted to local security considerations, e.g. amd
and local copying could be used.

12:30	Clive King (Dept Computer Science, University of Wales, Aberystwyth)

Filesystem change - a strategy to maintain user satisfaction!
*************************************************************

The user is the person that a system administrator should work towards
satisfying. A user should be allowed to use a UNIX machine with minimal
knowledge about the networks structure and without the need to cope
with of change. One of the main aims in managing a heterogenous network
should be to give to present a consistent environment to the user both
across the network and over time while allowing the administrator to
manage network change in a structured manner.

Change, as part of the evolution of a network can be the result of the
ever changing needs of the user, vendor driven with the new releases of
operating systems and third party software being installed in diverse
locations that many users are unaware of. A policy and mechanism is
required to allow the administrator to provide a uniform image of the
filesystem irrespective of where on the network it is accessed from.
One solution is to use symbolic links and an automounter to construct
the environment the user is familiar with.

With the large numbers of symbolic links required to give referential
transparency yet to facilitate management of non-system software, it is
clearly not feasible to generate all links required by hand.

This paper addresses the problem of providing referential transparency
in regard to system change and suggests a solution where a generator
for symbolic links takes a high level description of the links to be
constructed. The link generator used in combination with the AMD
automounter gives the user a consistent view of all filesystems across
the network.

Some of the design issues of the symbolic link generator are considered
and some thoughts are presented on a number of issues involved in
changing the organisation of a filesystem while minimising the impact
on the user population. Finally the issues and techniques presented are
placed in the context of he recent POSIX P1003.7.2 draft standard for
system administration procedures.


13:00	LUNCH

14:00	Lindsay Marshall (Dept of Computer Science, University of Newcastle)
	and Peg Schafer (Bolt, Beranek & Newman Inc, Board Member of SAGE/USA) 

SAGE/UK
+++++++

It is not only the working environment that is changing, the very role
of the System Administrator is changing. System Administrators are now
being recognised as being more than just skilled operators.  They need
to be involved in the planning process when new systems and networks
are being ordered.  However, at the same time organisations often make
inexperienced people take on the job of System Administrator without
adequate training or support.

In order to help raise the standard of System Administration and to
advance it as a profession, this conference will see the creation of
SAGE/UK - a UKUUG Special Technical Group for System Administrators.
Successful SAGE groups are already running in the USA and Australia and
the UK group will keep in close contact with their activities. The
group will provide a focal point for System Administration activities
and will organise workshops and tutorials as well as developing
guidelines for the proper management of systems.

15:20	Graham Dean (Dept Computing, Lancaster University)

The Devolution of Management
****************************

Many system management approaches falter at the first step due to the
fact that no explicit acknowledgement is made for the need for
cooperation and communication amongst managers and users. In many
institutions there is indeed a move towards the devolution of
management, with normal users taking on some of the tasks normally
associated with a system administrator such as the installation of new
software. To allow this maverick form of behavior to lie comfortably
alongside the constraints imposed by management strategies we must
provide a framework which supports the communication required to
resolve any difficulties found.


The approach I propose is to use an explicit representation of the
system to aid in supporting the communication required for effective
management. Through the use of a representation it is possible to
elicit information about normal working practices which can
correspondly be captured within the model.  This allows the easier
formulation of management policies and strategies and can also act as a
supporting knowledge base for system adminstrators.

15:50	BREAK

16:05	David Lomas (Academic Information Services, University of Salford).

What about the Workers!
***********************

System administrators fulfil a key role in the delivery of a computing
service. Some administrators consider their role is to master the
complex relationships of the software and hardware that comprise their
system. However, these computer systems and their administrators do not
exist in a vacuum Rather, these systems provide services to users
(workers) in their organisation.

This paper identifies some of the forces for change and the typical
responses to change from the perspective of computing service users.
The aim of this discussion is to remind system administrators of their
wider responsibilities to the organisations that they serve and how
administrators can address the concerns of their users.


16:35	Adrian Pell (Hewlett-Packard Laboratories, Bristol)

So I've got the data ... what now?
**********************************

Recent trends in computer systems have increased, without measure, the
problems encountered by system administrators. In particular, the move
from mainframes to distributed systems has contributed in two
significant ways. There are simply many more machines to manage, each
as complex in its software design as its large predecessors. In
addition, and perhaps of greater concern, the spread of distributed
systems has led to many client/server applications which are
fundamentally more difficult to manage because of their widespread
nature.


This paper present a methodology for describing aspects of computer
systems, both their hardware and software, in a way that permits a
product to be managed through its entire lifetime, from installation to
final removal. A description language is introduced, and first
experimental results from the Dolphin system are given.

17:05	Robert Evans (Centre for Computing Services, Imperial College)

User Management in the Imperial College Distributed Computing Environment
*************************************************************************

In 1989, Imperial College started to replace its central mainframe with
a number of Unix workstations situated throughout the college. It was a
requirement of the funding body that user administration remain
central. This also eases the maintenance of a single mail domain which
is used across college.  The number of users is about 8,000 and the
number of workstations in the college is about 600. These are spread
over three main sites, Silwood Park (Ascot), St Mary's Hospital Medical
School (Paddington), and the South Kensington campus.


A central server process was installed on a workstation within the
Centre for Computing Services. This process can be interrogated by the
remote workstation clusters which use it as a source for their password
files. Modification of the central database can be done not only by
centre staff, but also by authorized staff within the departments using
a suitably validated interface.


Each departmental cluster acesses the central database, usually
overnight, and builds a NIS database from this data. The cluster is
then run as a stand alone NIS domain with its ownserver. New users are
created on a cluster when their entry is in the password file, but no
home directory exists for them.

Password changes, and all other database changes, are made through the
central server. As passwords are updated through the central database,
there can be delays before the password is updated on the local
cluster. The process of scanning the password file for new users is
rather slow, hence this is usually done overnight. This restricts the
speed at which new users can be added. There is currently no policy for
how to remove a user!

The program grew, rather than was designed. A traditional problem which
can only really be solved by a re-write, possibly in C++. Various other
features have been requested. For example, password expiration, user
deletion, faster password updating, and better searching tools.


17:35	FINISH of System Admin Meeting


17:45	UKUUG E.G.M.
	(Details and calling notice dispatched separately to members).


+++ENDS+++





-- 
Neil Todd                       | ..In general, it is best to assume that the
PSI%234237100122::neil          | network is filled with malevolent entities
neil@pio.gid.co.uk              | that will send in packets designed to have
GID Ltd                         | the worst possible effect...	RFC 1122




-- 
Neil Todd                       | ..In general, it is best to assume that the
PSI%234237100122::neil          | network is filled with malevolent entities
neil@pio.gid.co.uk              | that will send in packets designed to have
GID Ltd                         | the worst possible effect...	RFC 1122