Suppliers would love the network manager to throw money at every
problem on the network. But couldn't you get the current set-up to
perform better? asks Nick Booth
If any good came out of the entire year 2000 fiasco it was that it
forced companies to take stock of the inventory of hardware and
software on their networks. After all, it's a lot easier to "fight
fires" when you know where the danger areas are located.
In fact, a network manager's job shouldn't actually be this
reactive, it just works out that way. By having a more
comprehensive knowledge of the organisation's IT infrastructure,
the network manager may well be able to compensate for a depleted
IT budget by getting the existing network to work harder.
There are two ways in which the effectiveness of the network can
be increased. By changing the way applications use the network, you
can reduce the burden on it, which in turn reduces the amount of
traffic it has to carry. On the other hand, you can soup up the
performance of the network by either adding more hardware and
bandwidth, or tuning your existing infrastructure. Post Y2K, many
network managers are finding the funds for expansion just aren't
there. The only option then is to redesign what you've got.
"If you want to change the way applications make use of the
network, you'll have to re-architect the bandwidth-hungry
applications," says Darren Prince, technical manager for network
performance supplier Smarts.
This isn't always practical, as it can involve a lot of work and
disruption while the software is re-configured or ported elsewhere
on the network. The time lost due to disruption can be more
expensive than adding to the network.
The quality of service mechanisms on a network's routers and
switches are classically under-utilised, says Prince, possibly
because this was a secondary consideration when these devices were
installed. The only other method of controlling the way
applications flood the network with their data packets is to use
dedicated traffic-shaping devices. Companies such as Xedia sell
devices that will do this very efficiently; but they cost several
thousand pounds and many network managers are told to improve
performance without spending any money.
If applications make demands on network bandwidth, the
alternative strategy is to increase the supply. This means
identifying the weak links on the network. Companies such as Jyra
Research will run one-off network checks, sending intelligent
agents out to identify which devices are slowing the network down.
This will tell you whether you need faster CPUs, more memory or
faster backplanes on your servers or switches. Upgrading items like
these will not break the bank.
It is not always a bad idea to throw money at a problem. The low
cost of bandwidth on a local area network will often make it more
worthwhile to install new equipment rather than allocate precious
man hours re-configuring it. To some extent, it could be more cost
effective to spend money to solve a problem in the local area,
because it won't cost much, and could be a lot quicker.
The return on investment for IT equipment diminishes in
proportion to the amount of kit you buy. Besides which, networks
are like motorways; as soon as you build one, users find ways of
filling them up. Any additions you make are expensive, and soon
become clogged with extra traffic. "Bandwidth-hungry applications
will still hog all the available bandwidth, no matter how much you
widen the pipes," says Prince.
The key then, is to devise a policy, not just for the way users
access the network, but for the configuration and load balancing
rules devised too. Many suppliers won't tell you this, because it's
in their interest for you to keep issuing cheques for more and more
hardware, but the majority of networks in the UK are acutely
under-utilised.
However, they do get used. A single network is called on to do
all kinds of things the original infrastructure wasn't designed
for. They have to support an increasingly diverse set of
applications and services, all of which have been added
incrementally. Users always want everything tomorrow, which is why
network managers shouldn't be blamed if they had an attitude of
"bung it on now and worry about it later". The problem is there are
now users with very different needs, from real-time transaction
processing applications to background Web browsing to multimedia
conferencing, all of which have an impact on the business at
different times.
"The issue facing network managers is not whether the network is
up, but whether the right services are available to the right users
at an acceptable level of performance," says Charles Muirhead,
founder of Orchestream.
Muirhead developed Orchestream in reaction to the problems of
the modern network, and being able to prioritise networks. Though a
supplier of "policy management" tools, Orchestream nevertheless
acknowledges that most IT budgets are already stretched to the
limit.
"There's a huge disparity between what companies want their
networks to do and the amount they plan on spending to achieve
those results. So network provision is no longer an acceptable
solution for service level demands," says Muirhead.
According to market analyst Gartner, networking costs are
typically 100% over budget. As this article proposes you tackle
your problems without spending any more money, we will overlook the
temptation to spend money on a network policy management tool (such
as Orchestream's) in favour of devising a policy of deploying the
existing manpower and equipment more effectively.
The first problem you should tackle, advises Phil Tee, founder
of network performance management specialist Riversoft, is to check
the configuration of products that have lain untouched since they
were installed. "It is very common to find that comms devices,
especially across the wide area, are on default settings or haven't
been updated," says Tee. "You often find routers aren't configured
to choose the cheapest lines to send data down. The most common
cause of pouring money down the drain is ISDN; the number of people
who've no idea how much their failover system of ISDN backup is
racking up their bills is amazing."
This is understandable really, given the amount of time that
configuring a router can take. Some people, says Tee, will waste
days trying to find the right combination of settings. "That is why
you should try and get the people who sell you the device to
pre-configure it before they send it to you," says Tee, "It'll save
you a lot of time and money if you get the supplier to do it."
Frequently, the supplier will be happy to do this as part of their
pre or post sales support service.
If you are going to spend a small fraction of your budget, says
Tee, it is best to spend it on getting your network tuned up. If
your network is firing on all cylinders, you're less likely to keep
fuelling the demand for more bandwidth.
If misconfigured devices are the biggest cause of wasted
resources, then lack of understanding of the needs of applications
is not far behind. Every application is mission critical these
days. Why install an application if it's not vital to the business?
The trick is to be aware of their tolerances for latency. How can
these users be kept waiting? When voice is sent over the data
network the figure is 20 milliseconds. When it's e-mail messages,
the figure is measured in minutes.
There are plenty of tools on the market that will allow you to
prioritise traffic. The most recent tools to be launched are
switches that examine the content of each data packet and decide,
according to pre-determined rules governing different applications,
what level of priority that packet gets in the traffic on the
network. Time sensitive applications, like SAP or real time
enquiries (such as an online purchase) are guaranteed bandwidth at
all times. Whatever is left is shared among the other
applications.
Security is a vital consideration but, in the course of a
security policy, sometimes networks can do more harm than good.
Firewalls, according to Darrell Woodward, security product manager
at systems integrator Wick Hill, can slow down proceedings to the
point where users give up using the system.
Most firewalls have a reporting mechanism that allows
performance information to be generated. Many people make the
mistake of assuming this is purely to tell them about attempted
security breaches. They're wasting a powerful management tool, says
Woodward. "It is important to read the firewall reports because
they'll show you how your bandwidth is being eaten up. Reports will
highlight the heavy users and the busy times of the day."
A major issue many network managers have to deal with is user
culture. Many end-users start the day by reading their social
e-mails and then responding to them, before embarking on their
work. This would explain the pattern most companies experience,
where e-mail usage is heaviest between 9am and 11am. There will be
another surge after 2pm after which Web usage is at its
heaviest.
Some companies legislate for this by allocating a maximum
2Mbytes for e-mail, the rest of the bandwidth being dedicated to
business communications. Other companies have dedicated PCs for Web
surfing, while others dedicate all bandwidth between 1pm and 2pm to
Web surfing, which encourages users to do their personal surfing in
their own time.
A heavy-handed approach to eliminating Web surfing will be
counter productive. When more and more companies are looking to
outsource their applications, and by-pass the IT manager, the worst
thing the IT manager could possibly do is upset the users by
getting bolshie about bandwidth usage.
A gentler method to encourage users to spend less time clogging
up the network with their Web searches involves creative use of
cache memory. "You may discover from your usage reports that a lot
of people download something like Acrobat reader or a particularly
large file from the Web," says Woodward. "The logical way around
this is to provide it locally."
Top tips for beefing up your network
- Find out a typical pattern of usage between
departments
- Set up a special surfing area
- Voice compression technology will allow four times as much
traffic to use the same bandwidth
- Utilise your existing bandwidth by sending voice by day and
data by night over the same wires
- Separate networks for storage, such as storage area networks or
network-attached storage can isolate the traffic as stored files
are retrieved from a single location
- Re-organise your applications
- Prioritise traffic using quality of service options on
switches
- Upgrade poorly performing devices (such as server CPU,
memory)