There is good reason for businesses to worry about the languages their IT systems use. Older languages, such as Cobol, are still heavily in use in vital applications, yet the number of programmers able to work in these "legacy languages" is shrinking rapidly.

Anyone with only a passing interest in the arcane world of Cobol and other legacy languages could be forgiven for thinking they had been largely replaced in the lead-up to 2000 as part of the "Y2K" panic.

But these older languages still account for between 12 and 15 per cent of new development - and it is concentrated in crucial back-end financial systems, explains Jim Duggan, a research vice-president at Gartner.

Meanwhile, the number of legacy programmers is down to about a third of the number in the 1990s, as studying old languages has been falling out of favour for many years and older programmers are nearing retirement - with little financial incentive to keep working.

"Everyone could go out and claim a pension today. The additional pension they would gain from working another year is small," says Mr Duggan.

Today's preference for newer "object-oriented" languages, such as Java and C# means that the older languages, including Cobol and Fortran and known as "procedural" languages, are rarely offered in universities and colleges, as students' demand for them is low.

"The demand coming from enrollees is definitely not for legacy," says Dave Arcemont, VP of global workforce development at EDS, the outsourcer that employs 39,000 developers. "When we assess graduates, legacy is obviously an area in which they're not adequately skilled."

Mr Arcemont says that between 60 and 70 per cent of EDS's developers have legacy skills, and demand from clients has led the company not only to focus heavily on internal training, but to work with universities in its home state of Texas to introduce legacy skills to the curriculum.

Although labour markets vary - Cobol programmers might go begging in one city and be desperately sought after in another - the figures are frightening for anyone responsible for running a crucial legacy system. Compounding the problem for employers is the fact that older programmers inevitably expect higher salaries.

Sixty per cent of mainframe programmers and adminstrators are 50 or older, compared with 20 per cent of those in Unix, Windows and Linux, according to a Gartner report published in November 2005. Another scary estimate widely quoted in the IT industry states that 50 per cent of those with mainframe skills will be eligible for retirement by 2007.

As many global business transactions still rely on legacy languages, this should be very worrying. Gartner also points to the many new developments being written in Cobol, an assertion backed up by a recent reader survey by Computerworld, the IT trade magazine, which found that 58 per cent of respondents from organisations using Cobol said they were developing new applications in that language.

With the growing staffing and expertise challenges, why do people stick with legacy systems?

Paul Green, a senior technical consultant at Stratus, has seen first-hand why these decisions are made. Stratus is a US-based company that sells a fault-tolerant operating system called VOS which is written in PL/I, another legacy programming language. Developed in the 1980s, VOS runs on Stratus's own hardware and is popular with financial institutions.

Although Stratus introduced support for Windows and, more recently, Linux, the VOS operating system is still widely used. More than half of Japan's credit/debit card transactions run on VOS and it has several very large financial services users in Europe and the US.

Mr Green, an IT veteran, says skills availability is a "tier one issue" for his customers when assessing whether to upgrade, but that alone is not enough to justify outright replacement.

"If any of our customers thought they'd replace their core applications, that's something they'd easily spend four or five years on - the first couple of years just planning," says Mr Green, who recently helped a large credit card company transfer from VOS to Linux on Stratus servers.

"The ironic thing is the new software is easily 100 times less efficient," he says, illustrating the breadth of views about legacy software. Some programmers admire the elegance and robustness of applications based on procedural languages; others despise them for their lack of flexibility.

Even those who are keen to move away from legacy systems altogether agree the decision cannot be taken lightly. Andy Reeves, chief information officer of P&O Ferries, the shipping company, has seen how a replacement can go wrong.

When Mr Reeves became CIO, an earlier attempt to replace the P&O Ferries' systems written in Fortran and Cobol and running on VMS Alpha, had run over time and over budget.

"By the time it got to the point where the software was being delivered, it had moved on so far the software was no longer relevant," he says.

Despite this, Mr Reeves still wants to replace the legacy system eventually. "The cost of ownership is horrendous. We put a fuel surcharge on the North Sea routes earlier this year and what should have taken a week in terms of coding took nearly three months," he says. "We have about four Fortran programmers who we rely on completely. It's hard to justify a team at around a quarter of a million pounds a year doing small patches."

It was decided to tackle the replacement project in smaller, safer chunks; moving it into a service-oriented architecture (SOA) using Cape Clear's ESB.

Many businesses opt for the best of both worlds: keeping some of their core legacy appli-cations, but running themwithin a more modern SOA architecture that makes maintenance and development easier and allows newer services to be added.

Migrating decades-old processes from old mainframes to modern platforms can prove difficult, however. This is where vendors of "clone" products - new programs that replicate legacy versions - have a role.

Sun Microsystems and Fujitsu provide these types of products, but the main vendor not linked to a server maker is Micro Focus International, a US-based company with roots in Cobol.

Micro Focus clones aspects of the mainframe operating environments to reduce dependency on the mainframe itself for tasks such as development and maintenance, and offshoring in small chunks.

Stephen Kelly, chief executive, says that there are still "a lot of Cobol developers out there", but that Micro Focus is also talking to colleges and universities in the US and UK about sponsoring short Cobol conversion courses.

The endurance of legacy code highlights the surprise comeback of the computers on which they run: mainframes. These big,proprietary computers were considered doomed when in the 1990s distributed computing and client/server models promised cheaper hardware and cheaper operating systems, such as Linux and Windows.

Happily for IBM, which dominates the mainframe market, reports of the mainframe's death proved premature and its business has enjoyed a broad resurgence since 2000.

Branching out into lower-end mainframe models and supporting Linux has undoubtedly helped. Bob Hoey, worldwide VP for System z9 Series sales at IBM, says that many customers find that the lower initial costs of a distributed system are outweighed by higher staffing costs required to maintain a group of single-application servers.

At the same time, the ageing legacy workforce is "one of the biggest concerns from customers worldwide," says Mr Hoey.

IBM, too, is tackling education. In 2004 it launched a five-year programme aimed at getting 20,000 people trained in mainframe administration, and recently embarked on another programme to simplify mainframe administration.

It aims to cut the training time for mainframe proficiency from two years to six months.