In the move from mainframes to a distributed system financial
services company Anfis dived in with both feet. Liz Warren has been
finding out how it worked ou
Distributed systems offer significant business benefits, but
they also bring a whole new set of management headaches. When Anfis
(Abbey National Finance and Investment Services), the central
services firm that supports the different brands within Abbey
National's Life Division, introduced client/ server-based systems
for the 1,800 staff at its head office in the mid-1990s, it found
it much harder to deliver the quality of service it had come to
expect from its mainframe operations.
"We didn't have the visibility we had in the mainframe
environment to manage servers, desktops or the network," explains
Scott McElney, IT service delivery manager at Anfis. "We would know
that we had a problem, but we wouldn't know exactly which server or
router was causing that problem. The only way we could find out was
to work methodically through our systems until we found it. That
meant we were taking far too long to resolve these issues."
These difficulties were particularly frustrating for the IT
service delivery team at Anfis because it had a strong background
in using systems management tools in its mainframe operations to
provide high quality and high availability services to users. What
it wanted to be able to do was to take its mainframe automation
methodologies and plug them into the client/server environment but
the market for systems management tools for client/server
applications was still in its infancy. It wasn't until Fraser
Cowan, at that time IT service delivery manager and since promoted
to head of IT services, saw a presentation on Computer Associates'
Unicenter TNG at a conference that Anfis felt it had found a
product that could address its needs.
No-brainer decision
In fact, Unicenter TNG met its needs so well that Anfis felt it
didn't need to develop a detailed business case for the project. As
Cowan put it, the decision was a "no-brainer". "We simply had to
get it in," McElney confirms. "From our service automation
background on the mainframe we were very aware of our requirements
and the benefits we could get."
Yet, despite the obvious potential benefits of Unicenter TNG, it
was still a brave move to commit to it lock, stock and barrel: like
the other client/server management tools available at the time, it
was still a relatively immature product.
Of course, having realised that more sophisticated management
tools for distributed systems were finally becoming available, the
Anfis team did carry out an evaluation of other products, including
leading rival Tivoli. However, it chose to stick with Unicenter
TNG. "We felt Unicenter offered us everything we needed and, at the
time, we believed it would be a lot more cost effective than
Tivoli," McElney explains. "We also knew that another company in
Scotland had recently bought Tivoli and we felt they would have
already got the pick of the best Tivoli people. As the only
organisation in Scotland at that time with Unicenter TNG, we felt
we would be able to get the skillsets we needed more easily."
Given the Anfis team's mainframe systems management experience,
it had clear ideas about what it wanted to do with Unicenter TNG -
but it also recognised that it had very little knowledge of the
product and exactly how to use it to achieve its goals. Because
Unicenter TNG had only recently been launched, skills weren't
widely available in the market for Anfis to employ on a permanent
basis and, in any case, the team wanted to get the project moving
quickly so that it could improve the service to business users as
soon as possible. It therefore decided to bring in an external
consultancy which could provide an injection of knowledge to
kick-start the project.
Anfis chose ICL as its partner because it felt it provided by
far the best response to Anfis's invitation to tender. "Most of the
other responses were rather thin on the ground and talked a lot
about themselves but not about Unicenter and we felt they didn't
seem to understand Unicenter," McElney says. "ICL had a lot of
information about the product and how to set it up. At the time, it
appeared to be the only company that had detailed knowledge about
the product."
ICL became a part of the Anfis project for about eight months,
joining in January 1998. It supplied two technical consultants and
a project leader to work alongside the internal team, although the
main project management role was assumed by a member of the Anfis
team. The first step was for the ICL team to interview staff from
Anfis to determine what systems the company had in place and how it
might benefit from the various Unicenter TNG components. This
allowed ICL to produce an design outlining which Unicenter TNG
components should be installed, and how they should be implemented
in Anfis's environment to deliver the systems management services
Anfis was looking for.
Once this design had been agreed, ICL set up a proof-of-concept
lab at Anfis to test how those components would work with the
platforms Anfis was running. Graham Bennett, a senior consultant in
ICL's IT consultancy unit, who was the technical design manager for
the Anfis project, explains that the proof-of-concept lab allowed
ICL and Anfis to test whether the product actually delivered the
functions in its specification and to check whether it had an
adverse impact on system performance.
Another key benefit of the proof-of-concept lab was the
opportunity it provided for skills transfer. The Anfis team worked
alongside the consultants from ICL, carrying out some of the
installation and configuration work under their supervision. On top
of that, as the ICL staff installed the Unicenter software in the
proof-of-concept lab, they were able to produce and test the build
scripts for each component and system. These provided a written
record of the work in the lab and could be used by the Anfis staff
once they came to roll Unicenter out to their live systems.
"ICL, produced a lot of thorough documentation which helped
achieve the necessary skills transfer," McElney says. "In
particular, there were some tasks which the ICL team had mostly
carried out and where we'd not really been involved, but at least
we knew what had been done because we had the documentation."
Once the solution had been thoroughly checked out in the lab,
ICL supported Anfis while it piloted it in one area of the
business. Working alone, Anfis has since successfully applied
Unicenter to a second area of its operations, as well as working
hard to develop and customise the solution further.
Throughout the implementation, the Anfis staff worked on
developing appropriate systems management processes and procedures
for its distributed systems, drawing on the techniques used in its
mainframe operations. "Our mainframe procedures were already best
practice," McElney says, "and we have found we can translate that
best practice to our distributed systems and work in much the same
way."
However, the process is not as straightforward as it sounds.
McElney points out that a mainframe produces very well-structured
messages, but there are no standards for message structures and
content - such as object IDs - across the various distributed
technologies and distributed systems often generate very long
messages where the crucial information is buried among a mountain
of other data. A lot of work is needed to make sense of the
messages generated by each platform and merge them into a single
interface.
This was not the only headache Anfis faced. The very size of its
implementation became an issue and one lesson Anfis has brought
away from the project - which, McElney says, has been confirmed by
other Unicenter users - is that it should not have tried to
implement so many modules at once. "Don't plan to do more than six
months work at a time," McElney warns, "otherwise it can become
like painting the Forth Bridge."
From ICL's perspective, Bennett admits that the major difficulty
was the lack of a formalised method for deploying the product,
since the Anfis project was one of the first Unicenter TNG
installations ICL had undertaken. ICL has since used its
experiences of this project to define a more structured approach to
implementation.
Technically speaking, the only major headache occurred with the
OS/2 system agent from Computer Associates (CA). McElney points out
that OS/2 was at that time already becoming obsolete, but he feels
the agent offered by CA simply "wasn't as reliable as it should
have been and didn't provide the functionality we needed." Anfis
ended up integrating a third-party product into Unicenter to handle
its OS/2 needs. However, McElney believes the experience shows
that, even with a comprehensive package like Unicenter TNG, you
need to be prepared to use third-party solutions in some areas,
because it is likely that some of your framework supplier's modules
will not be up to scratch. On the plus side, if you have picked one
of the leading frameworks, most of the third-party systems tools
suppliers will have already incorporated the necessary integration
in their products.
This breadth of coverage is one of the key benefits of a systems
management tool like Unicenter. Another is that, once the company's
policies have been embedded in it, the software ensures they are
enforced and that the correct procedures are followed.
McElney admits that it has sometimes been hard to get the strict
process discipline which is commonplace in the mainframe
environment accepted into the less structured working practices
that have grown up in the distributed systems world. "One of the
challenges we still have is to leverage a lot of existing system
management policies and procedures into actual practice," he says.
"Not everyone is on board with implementing what we want to achieve
in systems management terms into their policies and practices."
This cultural change extends down to the day-to-day working
practices of support staff. "Mainframe operators are used to
console operations where all messages come to a single screen," he
explains, "but it took some time to get across to support staff who
are used to looking after distributed systems and networks, that
all messages would be sent to their consoles.
"We had to convince them that they should be looking at their
consoles constantly to find out what's happening - not just for
notification of problems but also for proactive information they
can use to prevent problems arising in the first place - and that
they need to keep their pagers with them at all times so that they
can receive alerts from the system while they're away from their
desks. We also needed to get them to understand that they could use
the Unicenter console to diagnose and resolve problems without
necessarily having to visit the piece of kit which was at
fault."
Anfis was also quick to recognise that the implementation
project wasn't over once the software was installed. "The stage
that is the easiest to avoid doing but which provides the greatest
business benefits is the ongoing exploitation of the product once
it's been implemented," ICL's Bennett says. "A lot of people see
the implementation as the goal, but you won't get huge benefits
unless you continue to tailor the systems management solution,
because the out-of-the-box product is quite limited." For instance,
you should analyse data about systems issues occurring in the
previous month and determine whether there were any indicators to
suggest that a systems crash was about to take place. You should
then look at whether your systems management solution can be
reconfigured to detect those warning signs and alert you that a
crash might be about to happen.
McElney agrees, "It's really a continuous customisation process.
We are always looking for areas where we can improve our Unicenter
implementation and are looking for third-party products to plug
gaps where we can't find a Unicenter product that's good enough or
works in the way we want."
For instance, Anfis is now working to develop support for
layered automation. The system was originally set up, McElney
explains, to page the relevant engineer when particular system
messages were received. Now, when a message requiring action is
received, Unicenter can automatically initiate a predefined action.
If this resolves the problem, the message is cleared. Only if the
problem persists will the engineer be notified.
"We can't automate everything, but where we can, we will,"
McElney explains. "For example, we had one system which sent a
message to Unicenter that sometimes meant the system was down - but
not always. If the system was down, it had to be rebooted and, in
the meantime, Unicenter was automatically firing off an e-mail to
the business saying the system had failed. Before we automated the
process, it could take 40 to 50 minutes to resolve the problem. Now
Unicenter pages the engineer, who checks the status of the system.
If the server is down, he can press a single button which reboots
the server and sends a message to business users letting them know
that the server will be down for 10 minutes while it reboots. If
the system is OK, he can press another button which clears the
message and e-mails users to let them know the system is OK. We've
been able to cut downtime to just 10 minutes and keep the business
better informed."
Anfis has also recently installed the Web management option and
is now looking to bring its mainframe and more elements of its
network under the Unicenter umbrella. "We still have limited
visibility of our network and we would like to improve that,"
McElney admits. "Also, CA is coming up with new modules all the
time that we might implement."
In fact, McElney believes Anfis will be developing its use of
Unicenter TNG for many years to come, as the support team develops
new practices for existing systems and learns to deal with new
products as they are added to the infrastructure. However, one area
where McElney feels Anfis has gone as far as it can with Unicenter
TNG is with the anti-virus option. "It's a good product, which
gives us confidence in its detection abilities, and we have worked
on it constantly until we feel we have achieved closure apart from
various upgrades," he explains. "For instance, we have set the
server up to collect the latest virus signature files automatically
at scheduled times and to fetch new signature files in reaction to
e-mail alerts from CA. That meant that, for example, we had the
update for the 'I Love You' virus loaded onto our machines by the
afternoon of the day it hit."
It is this speed and level of control which lies at the heart of
the benefits Unicenter TNG delivers to Anfis, although McElney
points out that it's very hard to quantify the benefits from a
financial perspective. He suggests it's more about avoiding
disruption and downtime to the business, while minimising the
number of support staff needed or being able to deploy them on
tasks that add value to the business, as it is about making savings
to the existing budget.
"Unicenter is our eyes and ears and we can't live without it,"
he says. "If we took it out, we would rapidly be in a mess, with
more downtime and more extended downtime. That would affect the
service we could offer to our customers, so it would soon have an
impact on the company's bottom line. We have to ask people to
remember what it was like before we got Unicenter - and we were
running less complex systems then."
At a glance
The Organisation
Anfis is a central service company which provides resourcing and
facilities to support the different brands that make up Abbey
National's Life Division portfolio, including Scottish Mutual
Assurance, Abbey National Life and Pegasus. By sharing services
such as actuarial and technical advice, human resources, finance,
IT systems and customer services the various brands in the group
benefit from increased efficiency and operational costs which are
amongst the lowest in the life assurance industry. Anfis now
employs about 1,800 staff at its offices in Glasgow and is
responsible for managing about £20bn of client funds.
The Challenge
Anfis had installed a number of client/server systems to meet
business users' needs, but was finding it difficult to deliver the
kind of high quality of service it was achieving with its mainframe
operations. It needed a management solution that would give it
greater visibility of the elements in its distributed systems, so
that it could identify the cause of faults more quickly and, if
possible, solve them remotely.
The Solution
Computer Associates' Unicenter TNG offered a suite of modules
which could: present information about the various elements in
Anfis's distributed environment through a single console view;
allow support staff to diagnose faults and take appropriate actions
remotely; and control the actions they could take, to ensure
Anfis's processes and procedures were followed.
Secrets of systems management success
- Draw on the experiences and techniques of mainframe operations
when devising system management policies and procedures for
distributed systems
- Don't try to implement too many modules at once
- Use a proof-of-concept lab to check that the systems management
tools live up to their specification - and that they don't have a
negative impact on system performance
- If you use a consultancy to speed up the implementation, make
sure knowledge is transferred from the consultants to your in-house
staff
- Be prepared to integrate third-party management products in
some areas to get the depth or range of functionality you
want
- Automate as many processes as possible
- Don't underestimate the cultural changes involved in
introducing mainframe-style systems management to staff used to
supporting distributed systems
- Rolling out the system is only the beginning: the
out-of-the-box product is quite limited and you need to keep
developing your processes and procedures, and adding new levels of
automation and customisation, to get the long-term benefits of a
systems management tool
The Computer Weekly/Buy IT case studies offer an in-depth
analysis of a successful IT project, with expert comment from a
panel. BuyIT was launched in 1995 by the DTI and an alliance of top
industry bodies. BuyIT has selected best practice examples on a
range of projects. Each case study is scrutinised by the BuyIT team
of experts who make their recommendations and comments. The BuyIT
Computer Weekly Best Practice Series is endorsed by Fit for the
Future, a CBI-led, government-backed campaign to get business
learning from business.
What the BuyIT experts say...
Alistair Fulton
Chairman, BuyIT
Anfis is typical of a lot of companies which moved from
mainframe to distributed systems and found that maintaining
performance standards was more time consuming and expensive than
they had expected.
The central services company for Abbey National's Life Division
found the software to help with this problem but then realised it
lacked the skills and experience to tailor and implement it over
such a complicated IT network.
It was fortunate that Anfis found in ICL a company that could
combine experience of designing and implementing systems of this
kind with a good understanding of the financial sector. Had Anfis
not been so lucky, this would be a tale of woe, not a success
story.
We liked the approach ICL took in training Anfis' own employees
to recognise and deal effectively with problems rather than using
this as an opportunity to create further work for itself.
Anfis now has a system which consistently saves them time and
money. It is managing its network at a much greater level of detail
than before and its staff are sufficiently familiar with the system
to be able to handle future roll outs themselves.
I would be interested to hear from other companies which have
not had such good experience of the transition from mainframe to
distributed systems.
John Dutchman-Smith
Chairman of the IT Faculty, Institute of Chartered
Accountants in England & Wales
One of the key considerations I would like to have seen included
in the Anfis story is security. Companies have found they needed to
bring about a sea change in end-users' attitudes and approach to
security when moving to distributed computing.
Security requirements arise from three sources:
- Risks affecting the organisation's own information systems
which should be known by anyone experienced in those
systems
- Legal and regulatory requirements (including contractual
stipulations, and the principles)
- Procedures that an organisation has developed to support its
business operations.
The assets to be protected may take many forms. Principally for
information security, you will be concerned with information
assets. These may be supported by documents, software, physical
assets, people and services. In a distributed processing
environment these factors have to be considered at the level of
individual user departments and sections within the organisation.
Everyone has to understand that risk analysis is more concerned
with the cost of the loss of the use of the asset than with the
value of the asset itself.
One of the principal assets of any organisation is its image and
reputation. This may suffer when its information assets are lost or
corrupted. This, in turn, may occur when harm befalls any of the
other types of asset.
When processing is distributed to the end-users, the reputation
of the company depends on the end-users themselves.
Six steps to systems management heaven
ICL's method for implementing systems management solutions
involves the following six steps:
- Requirements analysis to determine, on the basis of the
customer's infrastructure and business needs, which products and
modules should be implemented
- Solution design based on that product set
- Proof-of-concept lab to test that the components can deliver
the expected functions and to determine the effect of the systems
management solution on the performance of the systems to which it
is being applied
- Initial user trial on a pilot site
- Roll out to the rest of the organisation
- Ongoing exploitation and development.
Feedback
Do you have any comments on this case study or any examples of
best practice of your own? Please send comments and input to
mark.lewis@rbi.co.uk
What did Anfis implement?