A Technical Strategy Report
This document summarizes recent discussions held among senior COs in the CS
department concerning priorities for technical developments over the next
year. Only areas requiring additional (or reduced) effort are discussed.
Other ongoing activities which are expected to continue are documented in the
activity figures and the project pages.
There was unanimous agreement that the following major areas required
significant immediate attention:
- Windows NT awareness and management.
- System management infrastructure.
- Infrastructure performance.
- Efficient and reliable Linux support.
- Efficient and reliable database support.
- QI replacement.
- Winding down of Macintosh support.
A number of other application areas were also discussed, and these are
itemized below. However, individual applications were considered less
crucial than the underlying infrastructure.
Investigation of Windows NT was considered a high priority for
several reasons, including:
- We believe that NT will be present in significant quantities in
the department within the next few years, and development of the necessary
expertise will take several years. It is crucial that we have
sufficient knowledge and infrastructure to avoid a repeat of the current
Macintosh situation where the linear nature of the support effort
means that we are unable to provide useful support for more than a
few machines.
- NT is already in use within the department in critical areas,
including the hardware laboratory, the database server, and several
research projects. There is a user-demand for more. These require support.
- Considerable effort is being wasted supporting existing NT
systems because of the lack of knowledge within the department.
- Macintosh support is becoming increasingly unviable (see below)
and we believe that NT offers the only foreseeable alternative. NT offers the only realistic possibility for providing several
applications, such as MS Office and graphics for slides and
the Web. This is likely to be the future standard OS for administration
and portables requiring these applications.
- More third-party applications (eg. Autocad) are now moving away
from Unix workstations towards NT as their primary platform.
- Many students (and staff?) will have NT machines at home, and we
would like to support the use of home machines as a way of reducing
pressure on departmental systems.
- The standardization on NT by other organizations, will make it
a pre-requisite for compatibility and collaborative work. This is
likely to be
true for some University applications as well (eg. FIS).
- NT may provide the most appropriate platform for first year
teaching, since desktop application software under Unix is woefully
inadequate, and NT may provide a more suitable Java development
environment.
Initially, we would like to increase our understanding of NT and learn
to manage the systems using some of the available standard tools. We
would also like to evaluate the applications under NT with a view to
using them as replacements for the Macintosh applications.
Eventually, we would hope to incorporate NT into our system management
procedures as they are developed.
Despite the claims of manufacturers, large numbers of machines still
require complex home-grown systems and procedures to manage in an
efficient way. This is true for all large sites, and especially for
academic environments where the systems are more diverse and evolve
more rapidly. Several years ago, a considerable CO effort was devoted
to developing a systems management infrastructure which enabled us to
manage large numbers of Suns very efficiently. This was extremely
successful and the department was acknowledged as a world-leader in
providing an excellent environment for research and teaching.
During the last few years, there has been a reduced number of COs and
an large increase in the variety of required support (Macintoshes,
Linux, the WWW, databases, security, PCs, portables, etc.). It is
this variety, rather than any increase in numbers which takes
additional effort, and the development and maintenance of the
infrastructure has virtually ceased. Support for Macintosh, NT and
(to some extent) Linux, is considerably less efficient than the
support for Suns (up to about ten times in some cases), and the
department is losing its reputation as a world leader in systems
support.This has produced a situation where the department may
perceive a lower level of support in some areas, despite increased
effort.
To reverse this situation requires a re-investment in the underlying
system management infrastructure, and we would like to start
investigating a re-design of several crucial areas, to take advantage
of experience and new technologies, and to incorporate new
difficulties, such as portable computing and diverse operating
systems. This includes technologies for machine configuration,
software distribution, and filesystems.
Some of this work is also a pre-requisite to any significant
expansion, such as further integration across the IPU.
With the increase in performance of desktop machines, the
implications for the servers and the network infrastructure are becoming
more noticeable. We propose:
- Installing a new Mail server.
- Installing a new News server.
- Increasing the memory and re-structuring the Web server.
- Spending some time monitoring network activity to determine any
possible problems or bottlenecks.
- Reviewing home directory and backup provision.
We believe that there will be a significant increase in the use of
Linux as the preferred Unix for desktop machines. The reasons for
this have already been well-documented, but include:
- More cost-effective hardware.
- Compatibility with home machines.
- Source availability.
- Dual-bootable machines (NT).
- Affordable Unix portables.
Our current Linux installation for CS2 is based on providing a
restricted environment. To make Linux available more widely on the
desktop (and satisfy the user-demand), we need to provide more
flexible automated configuration and a much fuller environment.
Most of the problems here can only be solved effectively by tackling
the general systems management infrastructure described above. Hence
we do not believe that we can support an increased population of
Linux desktop machines until the systems infrastructure problems have
been addressed. Very little Linux-specific work is required.
For the first time, the department now has a database technology which
should meet the database requirements of most departmental
applications. This is fully described in a separate paper, together
with the student record application. However
there are some dependencies that should be noted:
- The database server is based on NT and production-level support
for this service requires a reasonable understanding of NT management.
- Any new systems-management tools may well require a reliable
database as a core component.
We would also like to replace the QI directory with a real database.
QI is being used well beyond the original application (a telephone
directory) and is ill-understood, error-prone, and
difficult to manage.
Unfortunately (at least for those of us who like their Macintoshes),
it is increasingly hard to justify the use of these machines:
- The new OS (Rhapsody) has been re-positioned as a server operating
system and there are no plans to provide a desktop OS with memory
protection. This means that there will be no improvement in the
infuriating number of crashes and configuration problems.
- There will be no improvement in the managability of the OS,
which means that it will still cost an order of magnitude more effort to
manage than other systems.
- Future application support is uncertain.
- Support for our comparatively small number of Macintoshes is
currently costing almost two full-time COs as well as an unquantified
amount of academic staff time.
- MacOS runs only on a restricted range of (Apple) hardware.
To avoid attempting to support two operating systems simultaneously,
we propose switching (supported) administrative and portable computing
from Macintosh to NT in one move, if this still appears sensible once
we have investigated NT in more detail.
Solaris 2.6 provides support for NT and Macintosh
filesystems as well as some other features which may save effort
elsewhere and increase security. Solaris 2.6 is also an
essential requirement, should we wish to purchase more modern Suns.
We would like to investigate, and
possibly implement, this as soon as possible.
Security is still a growing concern. We still do not believe that the
department understands the true lack of security in the department
and the resources and decisions that are necessary to improve this.
We are aware that there is no serious Java expertise within the CO team
and we would like to develop this, since there is a growing demand,
which would increase if/when Java is used for CS1 teaching.
We are conscious that several new applications would be very useful,
and several existing applications badly need work. However, since
these tend to have few dependencies, and represent new areas of work,
it is often difficult to fit them into people's existing work.
- A network fax system.
- The ``faults'' program
is running under an obsolete language and is very difficult to maintain.
This needs significant work or possibly replacement.
- The door locks program is in a similar situation.
By Paul Anderson <paul@dcs.ed.ac.uk>
, Alastair Scobie <ajs@dcs.ed.ac.uk>
, George Ross <gdmr@dcs.ed.ac.uk>
Department of Computer Science, Unversity of Edinburgh, 16th October 1997