How to get started on Mainframes and is it worth while?

You might want to play a bit with the Hercules emulator from IBM zSeries and it's predecessors.

There's a few older releases of OSes like IBM/VM, OS/360, MVS available here.

If it's viable? Yes, even if the market share it's limited, is actually quite stable.

It's funny, this is really one of the most interesting questions I've read on SF and I've been thinking about my answer to it longer than I normally would, so here goes.

As Robert mentions people have been sounding the death knoll of mainframes since I was a kid and initial thoughts on the subject would be to run away from the area - there's definitely more client/server systems out there and this has marginalised jobs in mainframes.

That said the jobs in this field are usually pretty well paid compared to say linux-based jobs. The main reason is that it's typically large companies who buy mainframes and they're inherently mission-critical systems too. Some of the best paid contractors around are mainframe guys and I know just how high IBM's professional services costs for these systems can be.

One thing about mainframe support and development is that these systems are often under very strict change control with straightforward upgrades/patches etc. being planned over months and quarters. This means that support staff often have an unusually high degree of 'research time' available to them - i.e. get a job in this world and you'll have plenty of time to skill up in other areas if you have the motivation. Another benefit of working on these large, critical system is that of job security, sure these machines get replaced and people lose jobs but usually not without significant notice, it's not uncommon for a mainframe to take two years to replace in a business, giving you time to look around and/or skill up elsewhere.

Now if you're after the thrill of a start-up you'd be better going the linux route, it allows for very rapid prototyping and movement from proof of concept into production, but if you prefer a steady state environment I wouldn't shy away from mainframe work at all - as I say you could end up making more money anyway.

I hope this of some help.

Every fall IBM seems to offer a "Master the Mainframe" contest that might interest you. Registration is closed for 2010, but perhaps the documentation would still be interesting.