The death of the mainframe is about as elusive as the year of the Linux desktop. But cloud computing might finally present a terminal opportunity, so to speak, to those stalwarts of big business computing, by providing a compelling answer to the twin stories of reliability and throughput that have always been highlights of the big iron pitch.
Advocates of big iron talk about reliability. But with public clouds, we’re learning how to build services that achieve very high levels of reliability despite having low individual node reliability. It doesn’t matter if a single node in the cloud fails – cloud-style architectures route around that damage and keep the overall service available. Just as we dial storage reliability up or down by designing RAID arrays for the right balance of performance and resilience to failure, you can dial service reliability up or down in the cloud by allowing for redundancy. That comes at a price, of course, but the price of an extra 9 is substantially lower when you tackle it cloud-style than when you try and achieve it on a single piece of hardware.
The other big strength of big iron was always throughput. Customers will pay for it, so mainframe vendors were always happy to oblige them. But again, it’s hard to beat the throughput of a Hadoop cluster, and even harder to scale the throughput of a mainframe as cost-effectively as one can scale a private cloud infrastructure underneath Hadoop.
I’m not suggesting insurance companies will throw away their mainframes. They’re working, they’re paid for, so they’ll stick around. But the rapid adoption of cloud-based architectures is going to make it very difficult to consolidate future IT onto mainframes (something that happened in every prior generation) and is also going to reduce the incentive for doing so in the first place. After 20 years of imminent irrelevance, there’s finally a real reason to think their time is up.