Can't you get to that uptime with a well-configured cluster? I tried to find more information but couldn't find any other reasons why mainframes would still be the best option for a newly designed system?

Getting a cluster to guarantee that something happens excatly once is very hard, but doing things exactly once is an extremely useful property in banking. Using a single earthquake-proof computer with redundant everything, including hot-swappable RAM and CPU gives you similar reliability while making it much easier to achieve many desirable properties. Even if you can do the same in a cluster, spending more on hardware to reduce software complexity is worth it when bugs might cost you millions or billions of dollars.

Also, with Mainframe, you’re not only using a different machine but also a different “everything”. You need specialized storage, you need to use an “exotic” programming language, you need people who know JCL, RACF, DB2 etc., for a lot of “middleware” (CD pipeline, access management, version control) you have limited vendors to choose from so you’ll pay a lot for that too. So if you are also running something else (Linux servers etc.) you basically need double the staff and licensing costs...

That is not really 100% true. You can certainly run z/OS if you want, and that incurs those costs you mentioned - but that is true if you want to run Windows as well for example.

However, the biggest thing to remember is that you can and should run Linux on these things as well. Linux on z, or zLinux as it for some reason is called, is just Linux on the redundant and fault tolerant mainframe hardware. Anyone with Linux experience could manage it really, and you would get a pretty damn good platform to build a high availability service on.

> Node.js on z/OS provides extra security and performance by leveraging the capabilities of IBM Z.

https://developer.ibm.com/mainframe/2018/01/19/reasons-host-...

Good old days when I transit from XA to ESA, from dos/vse to movs, from racf administrator to dB admin, Ibm start to ship 3390-3, ispf, Rexx to generate jcl, patching and debug 370 assembler ...

Still holding a Hp dos pocket whilst working on all these.

ThOse were the days. Different very much from using a mac to run leela zero using egpu :-)

The easiest of there has to be DB2.

Author here, I might write a blog article about that because it is really quite interesting.

The TL;DR is that some might want to rather run 1 or 2 systems (mainframe) instead of 100 physical machines (conventional distributed cattle system).

Now, IBM does make it quite expensive but the mainframe has some pretty cool features like pay-for-what-you-use (which you of course get with the cloud, but not so much if you also want your data in-house).

Anyway, it is a fun beer topic if nothing else :-)

Well...you sort of “pay for what you use”. I believe the extremely simplified version is you pay for the peak CPU usage (excluding specialty processors like ZIIP) in a moving 4 hour average window.

Context: I work for a large organization that used to run several physical Z13 mainframes, all of them containing several sysplexes. If we had issues, an IBM consultant would fly in within the day. We were definitely not IBM’s biggest customer but we were not insignificant for them either.

We had a lot of mainframe support staff (so not people programming for mainframe but people maintaining storage, DB2, z/OS upgrades etc.) and I think even for them, the IBM bill was more or less: we see a large number, no idea why it’s this amount, but we cannot prove it is not right, so I guess we’ll just pay it.

Mainframe billing is really complicated.

It does sound interesting, so please do.

IIRC the mainframe uptimes are for the cluster („sysplex”) anyway.