|
|
|
Using commercial statistics programs in a small-company
environment
A business model for IS software
Word 97 version
1. Introduction
2. Solution
3. Business model
4. System
5. Accounting
6. Licences
7. Licence-based installations
8. Support-based installations
9. Seats and processors
10. Showcase
11. Marketing
- Introduction Next Top
The Industrial Statistics system is designed to assist companies use proper
statistical procedures in their work. In many instances, companies are so
small that they do not have the appropriate expertise or access to the best
software or, if they were to install it, the necessary IT resources. They
therefore cannot afford to buy in statistical expertise nor to install and
maintain the appropriate tools.
These are not only the tools that would be recognised by an academic statistician;
it is essential to include processes that are understood in an industrial
context such as Taguchi and 6-s . The applications
areas also need to be included as many topics like neural networks, pharmacokinetics,
forecasting all involve a large amount of statistics. These are often disciplines
where the wheel has been re-invented under another name but the underlying
concepts are recognisably statistical.
For a problem that ideally needs specialist software and statistical advice,
the cost of this software, the expertise required to use it and the IT maintenance
overheads mitigate strongly in favour of the quick-fix spreadsheet approach
– the modern equivalent of the back of an envelope. This may well
appear to solve the problem but of course such answers can be completely
wrong or at least misunderstood.
We should realise that, for example, a typical manufacturer in the UK will
have fewer than 20 employees – perhaps even fewer than 10; such companies
cannot afford expensive statistical consultancy at thousands of pounds a
week, let alone software which costs even more and the overhead of maintaining
it. Our system is designed not only to enable such services to be used by
small individual companies but also to foster clustering so that companies
and consultants can work together with ease. Thus the 20-employee manufacturing
company which is really working with other suppliers to its customers can
appear to work as one integrated unit.
By making statistical services available at a price the individual companies
can afford, the costs of design, collaboration, production etc can be brought
down into the range of acceptability, quality and reliability improved and
therefore profitability increased. This is true not only of manufacturing
companies but also of small high-expertise research companies, for example
in pharmaceuticals or life-sciences, as well as consultancies in many areas.
Another example of statistical need occurs in medical statistics; statisticians
have largely abandoned hospitals in favour of critical mass university departments.
Many research medical staff are therefore left without immediate statistical
expertise and software at a time when journals rightly demand statistical
skills to be applied throughout the research topic.
- Solution Next Top
We are addressing these problems at www.industrial-statistics.com by providing:
- A free registration and search facility so that potential users may
find statisticians, consulting companies, trainers, research institutes
and university departments best suited to their needs, together with help
mail lists and other professional tools, and
- A statistics server – effectively a specialist bureau service
- that enables all parties to work using the best or familiar software
in a collaborative mode.
The IS statistics server is an open (but not free) access statistical laboratory.
The aim is to have the latest and best of all software and facilities so
that clients can use the right – or familiar – tool for the
job. This includes a lot of open source software but we are keen to offer
closed source proprietary products that are robust and well known. Clients
frequently prefer such software where there is a clear line of responsibility
and demonstrable fitness for purpose.
- Business model Next Top
Enabling access to software has long been associated with problems of security.
There are various stakeholders to the process of using software in a collaborative
context, external to the industrial statistics site:
- The end user
The end user wants to be able to run a variety of professionally installed
and maintained analysis packages and to know that their data is secure
from prying eyes while being available to trusted colleagues and collaborators.
- Third party licencors
The software owner wishes to see an income stream without violating intellectual
property violated and to be able to control the software in some way or
another.
- The system administrator in client companies
The sysadmin tasks of offering external access securely will generally
require a complete rethink and reconfiguration of their system. This is
a task always avoided if at all possible. Mounting and maintaining a number
of packages that may be rarely used is another overhead.
If these problems can be solved, there are big advantages to using a common
platform. Again, by the principal stakeholders:
- The end user
The end user – as a statistician, collaborator, ultimate consumer
or trainer – will be able to use the appropriate and familiar tool
for the task at hand and see the results immediately on an owner-ship
neutral platform. It will not be a case of one party allowing access to
a possibly temporary collaborator on their own system with all the security
issues that such a move raises but access on a secure platform where they
can control visibility and also who pays the bill.
Such a system can promote collaboration and clustering and, given the
global market, at a transnational level, as well as driving proper procedures
into a fragmented industrial market.
- Third party licencors
The software owner has the chance to showcase their product without allowing
access to their own systems or risking intellectual property. As will
be seen below, this can also generate an additional income stream via
individual support agreements that can be fully policed by the software
owner (or the web-site management if required) with minimum effort and
that may grow into conventional licence sales. In addition, conventional
sales via downloads can be arranged either from the industrial statistics
site or by linking to the vendor’s own site.
- The system administrator in client companies
The system administrator’s task in the client companies is not affected
– other than ensuring that their computers can access the Internet
and maybe allowing some restricted ports open in their firewall for X-displays
etc.
- System Next Top
The fundamental basis for the server is Linux. This has been chosen for
reasons of robustness, flexibility and cost. The system currently runs a
version of RedHat 8.0 with a modified kernel, as detailed below. The initial
platform for the bureau is a robust professional rack system installed at
Manchester University. Further details of the configuration are available
on request to interested parties, subject to the usual confidentiality understandings.
The key to the security problems is a robust and secure patch to the underlying
kernel. There are a number of these available for the Linux kernel but,
for reasons that will be apparent later, the patch chosen is the grsecurity2
patch which includes role-based Access Control Lists with a sophisticated
learning mode. This enables only those parts of the system that are strictly
necessary to be visible to the user.
A number of services are available but the most important of these, from
a consumer point of view, is access to a fully functioning Windows 2003
Server which currently runs on a VMware Virtual PC itself hosted on the
Linux server. While we wish to enable Windows facilities, the extra cost
of running this needs to be taken into account. Where a program is available
under Linux, it is expected that a user will take advantage of the lower
running cost. Some programs are only available under Windows and, as long
as there is sufficient market and it is a suitable program in the judgement
of the management, it will be mounted.
Windows is just treated as a service with a common login user name (although
a different password may be used) and the remote drive is a Windows directory
under the user’s Linux user. The user’s Windows files therefore
enjoy the protection of the grsecurity system in the same way as do the
user’s Linux files and Windows security aspects are left to handle
installed software. (The Windows service is currently in beta and not yet
publicly available).
Other programs are also installed as a service. The user can apply to use
those programs for which the vendor wishes to retain some control. The vendor
has an adminsitrative user dedicated to their product and this contains
a file that has a list of authorised users. This file is scanned regularly
to ensure that the ACLs are kept up-to-date. Therefore a user is only able
to use a product if the vendor so agrees. If the vendor is happy for all
users to have access – perhaps because their licence terms allow it
and the appropriate fee has been paid – then the ACLs will automatically
include the program role in all user profiles.
It is essential to ensure that software licences recognise the type of
user that may use their software. The licence must permit commercial use
of subscribers to the bureau who are by definition not employees of Quantex
Research Ltd. Only with this full understanding is it possible to proceed
and this may require specific permission – particularly if they are
not in fact the prime software producers.
- Accounting Next Top
As part of the grsecurity package, an accounting system is installed that
measures the usage – however the program terminates – of specified
programs in CPU usage, IO, space and bandwidth. This enables differential
pricing to be applied so that the client only pays for what they use. This
ensures very fine granularity of control of who has what access to which
files and programs with minimal overheads.
The implementation works on the basis of an account where computation units
are purchased as required and the account debited by the usage on a day-by-day
basis. Rather like a bank, the system pays an effective ‘interest’
on the daily balance and provides an automatic discount scheme for large
purchases of facilities. It also includes ‘family’ accounts
where a number of users draw from the same account. Access to standard Unix
groups is also provided so those users outside the ‘family’
can still access private data within a directory. These are all designed
to encourage collaborative working.
In addition to consumption, it is also possible to add episodic or temporal
elements. These are particularly designed to ensure an income stream to
the software owner for facilities such as support and maintenance.
- Licences Next Top
Given that we have accurate process accounting and a sophisticated permissions
structure, there are two main approaches to mounting third-party software:
We can:
- Purchase the licence in full, recovering the cost from the users, or
- Make the software available on a per-user basis to those users that
have purchased support either directly from the supplier or via our services.
In the second case, the cost to Industrial-statistics.com is expected to
be small – in fact we would be providing a service to the supplier
and can form part of the marketing strategy for the product. We discuss
this feature in more detail below. Note that we do not provide support ourselves
but we are happy to set up a forum if requested.
In all cases, an administration account will be provided – sometimes
managed by us. This can enable the owners to install and update the product
themselves, depending on what access is required to the system. For example
a product placed in directory /usr/share/programx could have a directory
created of that name by the sysadmin and any other resource made available
so that the software company could do the installation and maintenance alone.
Such an account will of course have restricted privileges – other
programs will not be available and only testing of the installation will
be permitted. Reasonable use of the program from the installation user will
not count towards general program use or cost. It would be hoped that the
software provider, if they want to use the system to demonstrate their product
at all – perhaps even for training purposes - will have an ordinary
account which would not of course be able to modify the installation but
would have full standard access to all programs, subject to having registered
to use them.
The information provided to third party installation and maintenance accounts
is not very extensive – only a unique process identifier, start dates
and CPU times are provided. The sensible restrictions of the Information
Commissioner mean that individual users cannot be identified, nor of course
for commercial reasons can their work. Quantex Research Ltd is registered
with the Information Commissioner.
- Licence-based installations Next Top
Where only the licence fees are payable – ie there is no support
element purchased - this can be very expensive with no guarantee that the
software will be used sufficiently to offset this cost. Therefore we propose
four ways of funding such licences:
- Purchase the software licence ourselves, subject to access being allowed
in the way described above and recover the cost by appropriate charges
to those using the software, (this is the approach for all open source
software implemented),
- A closed group of users may purchase the software themselves, we can
install it on our system and the users can share the program subject to
some small management charges,
- The group as described above may also make the software available to
others, recovering the costs from those users external to the financing
group. This is a general case of 2 above.
- A single external company – typically the software producer –
may install the software and recover the costs in a similar manner. This
is a special case of 3 above.
All these can be configured as variations on the basic installation user/permissions
model outlined above.
Where we recover costs and pay the software producer, it will be possible
to include some anonymised information if needed – such as the country
from which the user has registered. This will enable the software owner
to distribute funds to the different parts of their own organisation. It
is important to recognise that, as a registered Company under the UK Information
Commissioner, we are obliged to ensure the privacy of our customers and
so the requirements of individual software producers must be both reasonable
and declared to the user when s/he registers to use the product.
The running costs and management precept for all cases will be negotiated
individually but will relate to the CPU consumption only – disk space,
IO and bandwidth are charged at a common rate for all programs installed
including shells and compilers. The running cost need not be a linear function
of the time – it is possible for example to make it quadratic or charge
more or less above a given break-point.
- Support-based installations Next Top
For some software producers, a substantial part of the licence fee is to
provide after-sales support. This cannot be provided to all and sundry that
may happen to use their software while logged on to the Industrial Statistics
server. Equally we cannot provide support for all the software mounted.
As indicated above, we provide two mechanisms that can ensure this income
stream.
- Make a time-based and/or episode-based charge and pass this on to the
software owner. These charges would have to be reasonable and there are
advantages to either time or episode based approaches.
- There may also be a case for a maximum and minimum to be applied for
any one family. Clearly the software producer would have to know who has
subscribed to such support, which could be done automatically when they
register or renew.
- Give the software owner not only an installation user but a user that
can act as an administrator that may authorise individual users to run
the software.
- This could be done, for example, by placing a simple text file of authorised
users in a directory that is scanned from time to time by a program that
sets the roles for each user. The control of this text file would be entirely
within the software producer’s hands, who could place a checksum
or other security systems to ensure that the file had not been altered.
The options are therefore either to leave it to us to provide the charging
mechanism, which will still need the qualified users to be identified, or
the software producer can handle it themselves, which means interfacing
with our server system in some way – that could of course be manual.
In either case, the authorisation could be at the individual user or family
level – charging via our system will always be at the family level.
There may be a case for a small or heavily penalised amount of unsupported
use without having to enter a support agreement, just as a taster.
- Seats and processors Next Top
Two important issues are seats and processors. Some conventional licences
restrict the number of people that can use a product at any one time and
the number of processors on which it may be used. This is perfectly understandable
even though the processor speeds are being increased all the time. It is
possible to use the batch system but this is not the best solution.
Such restrictions are very annoying to users where they have no managerial
control over access, such as in a bureau. Should a user try to run a program
and be stopped by the seat limit, they are likely to move to another program
if at all possible. Issues like multi-processors, clustering and hyperthreading
also arise. Since we intend to expand the system using a cluster as soon
as is necessary, this is of further concern.
However as there is a fully enabled accounting scheme in place, both these
issues are irrelevant – the more times a program is used, the more
money is earned. Seat and processor limits are therefore counter-productive
in such an environment. In a cluster environment, while it is possible to
limit a particular program to certain nodes only, this may also be counter-productive.
The aim surely of a software producer in the present context is to encourage
use of their software under all conditions and not to put impediments in
the way that mean either sluggish performance or not being able to start
a job at all.
We would therefore prefer that a licence under licence-based models ii-iv
in particular is relatively open and flexible. However it may be that a
software provider wishes to limit the use of their software precisely to
encourage licence sales – essentially becoming a demonstration installation.
This could be done on a per seat basis, a processor limit and/or by limiting
the resources that the program can use.
- Showcase Last Top
We have already pointed out that this is an opportunity to showcase products,
enabling potential customer to ‘try-before-they-buy’ or just
solve a particular problem without a large financial commitment. The client
and consultant may select the most appropriate software for use from a large
portfolio. We are happy to act as resellers of software if the appropriate
terms can be agreed.
The user, having tried the software, may prefer to purchase a licence rather
than depend on an outside agency. This may be due to a number of factors,
including:
- The user wishes to run simulation or other techniques that require a
lot of graphical interaction, and the latency of working over the internet
may become a problem,
- Having developed some core software, the user may wish to interface
the software with systems that we don’t run or make available –
eg databases which are difficult to control and monitor.
- Interaction with other systems such as in real-time forecasting could
make it be difficult for confidentiality and technical reasons, although
not impossible,
- The software is appropriate to their needs and heavily used.
These approaches will therefore also drive software sales. There is of
course the risk that some companies will choose to use software via the
bureau service rather than buying a licence. The balance between these two
states is a business decision on both sides and, given the alternative income
stream from support, it may be the better solution for all.
Where there is a substantial service income stream to the software producer
from support – either on a temporal or episodic basis, there are implications
for the business model. Industrial-statistics.com will be providing the
producer with:
- a platform, and
- access to a large number of industrial statisticians.
Hence this can form a part of the marketing vehicle for the product and
this is a service-based business model. There is a case for the producer
renting the facility from Industrial-statistics.com.
- Marketing Top
Industrial-statistics.com will be marketing this service widely to:
- Freelance consultant statisticians who do not have access to large institutional
facilities,
- Academic statisticians either who (correctly) need to use commercially
licenced software for external consultancy work or whose IT departments
are not willing to mount particular packages,
- Medical research staff who require competent statistical services and
software,
- Training companies that may locate clients by the search facilities
as well as offer the bureau service to course participants,
- Manufacturing companies,
- Other users such as some of the new and innovative pharmaceutical companies
or bio-science companies where there is no in-house statistical expertise
or software,
- Larger companies that may choose to use the family account systems,
- Employment agencies that may use the search system and suggest that
their clients take advantage of the server.
- Occasional or drop-in users.
The first stage of marketing is be directed towards statisticians and trainers
– we cannot offer a service if there are no potential service providers.
Much of this is being done by articles, seminar and conference presentations.
This will be followed by targeting manufacturers and pharmaceutical companies
generally via trade exhibitions but also by individual presentations. It
is possible that some conventional advertising will be done but the best
promotion is by word of mouth and encouraging collaboration.
All of these may be in the UK or elsewhere but the initial target will
be to a UK audience. Internet marketing techniques, search engine optimisation,
affiliate schemes etc will also be employed to promote the service to a
global market place, coupled with additional computation at strategic locations
to reduce latency problems.
John Logsdon
Quantex RESEARCH
j.logsdon@industrial-statistics.com
http://www.industrial-statistics.com
+44(0)161 445 4951
| |
|
Copyright © Quantex
RESEARCH Ltd
Last modified: 27/04/2004 22:00 Site build 17/06/2005 23:25
|
|
|
|