Following IBM's announcement of its new System z9 mainframe, which focused not so much on added performance but on security features, we asked IBM what prompted it to move away from the more traditional metrics and measurements. And of course we asked IBM's spokesman, UK server strategist Doug Neilson, if the mainframe was dead yet and, if not, why not.
Q: For years, people have been expecting the mainframe to fade away. What does it bring to the modern IT world of high-powered, highly capable servers?
A: Don't forget that many of the techniques that we use on a daily basis came from the mainframe. Mainframe technology has cascaded it down to the more familiar servers -- Intel and Unix servers. An example is virtualisation, which Intel and many others see as a very important technique right now. The mainframe has been using virtualisation since 1967. Also with storage, IBM virtualises the disks -- our virtual SAN server does that but mainframes have had that since 1972. The short answer is that we know how to talk to customers about it and what the value of the mainframe is.
We want to spread those benefits to the enterprise and make the mainframe an administrative server and backup for the enterprise.
Q: So if I can do all that on a standard server, why do I need a mainframe?
A: Other servers can run VMware -- say four, eight or 16 VMs. We can run thousands on the mainframe. We have customers who run 1,500 on one box. One even ran 97,000 copies of Linux -- just for fun. It's about capacity and reliability. We've had 200 new customers for mainframes in recent years. They want a reliable secure server that can scale and is manageable. What happens is that they come to us with their requirements and we provide what they need. They buy a big server then they realise that it's a mainframe.
The key issue is cost. Customers with distributed systems -- lots of boxes -- are starting to be honest about costs -- about management costs not acquisition costs. Some of their systems are very expensive. To take a typical metric, if you have one systems administrator per 30 systems, what do you do when your system grows to 3,000? It becomes very expensive. All those systems are like icebergs -- invisible and below the surface. And there are hidden costs as often, non-administrators tinker with them even though it's not their official job, and that's is a cost to the business.
With a scalable mainframe, it grows and the number of administrators does not increase. We think the pendulum is swinging back to centralisation -- it could be using Intel or Linux too -- but the mainframe has big part to play.
We are also trying to build an ecosystem around the mainframe. They are starting to teach mainframes in universities now, because we want them to understand them -- it's about hearts and minds. We need to get it out of the cupboard as it does the same stuff as many servers, like running Linux for example.
It also runs Java -- the z9 uses the only Java-dedicated microprocessors because we have to deal with vast numbers of JVMs -- up to 16 in the top end products. We introduced that in last 18 months.
Q: What made IBM focus on security with z9 above anything else?
A: The mainframe has always had security and it's a good time to talk about that right now -- there are a major requirements for both financial and privacy reasons. And some of the market may not know what the mainframe does in that realm.
We want to spread that technology across the infrastructure and the way we're going to do that is have the mainframe act as security manager for all platforms. For example, we would like to see data encrypted at all times. If you put a tape in the post, the data on it should be protected by encryption. To do that requires a lot of technology. It means you need hardware encryption support both for processing and for management of the cryptography.
Encryption is an ideal job for a mainframe. The z9 has dedicated, IBM-designed encryption processors within it. What's more, the crypto chip will self-destruct if you try to remove it in an unauthorised way -- it uses an acid capsule to dissolve the chip, I think. The machine has high levels of certification from the government, and is secure to near-military levels. That takes a lot of engineering.
Q: You say that the mainframe offers reliability but other vendors say that.
A: We contract to customers for a mean time between failure of 37 years. So others may be very reliable but we try to engineer out scheduled downtime which is 90 per cent of downtime. The Z series can be upgraded non-disruptively while it's running, including the OS, all without bringing down the system. So it's not just about hardware or software failure.
Q: What chip does the z9 use, and is it related to any other technologies?
A: The mainframe processor is a bespoke IBM design that's not available anywhere else. The top-end product in the z9 range uses 64-way symmetric multiprocessing, and you can build clusters for greater capacity, especially over a distance to add resilience. The architecture is unique. It's CISC so it's not like the [IBM's] Power5 chip -- if anything you could say the Power5 is a distant derivative of Z System microprocessor.
Q: Why do we not see standardised releases of information as with other server-type products?
A: We can't talk about what our customers are doing because the customers who are getting business value out of them, such as banks, supermarkets, governments and telecoms companies, like to keep them locked away. They don't talk about them. As for benchmarks, it's very hard to benchmark a mainframe because it runs so many processes -- that's how it integrates the business -- but it's near-impossible to measure. So a benchmark is not a realistic measure of performance because of the sophistication of the processes that run on a mainframe. Instead, we run workloads for customers and provide them with spreadsheets that customers can use for capacity planning.