Implementation for z/OS (IBM mainframe)

[andrew890]

I was interested to try this on z/OS i.e. IBM mainframe operating system, using both C++ and Java.

C++ required a few changes due to newer compiler features not present.
Java ran without change, however a .gitattributes file with

*.java zos-working-tree-encoding=ibm-1047

was required to checkout and compile the Java code on z/OS.

Are these changes worth contributing back to the original repository? The C++ version will run under Linux, but there's nothing new and no performance improvement in the code itself. Just the ability to run on a different platform.

[rbergen]

I'll start with this question:

Are these changes worth contributing back to the original repository?

In my opinion. no. This repo targets the x86-64 and aarch64 architectures (the latter to the extent possible), and the existing C++ and Java solutions work on both. As I also mentioned in response to this PR of yesterday, I would love to take a look at the details in a "show and tell" discussion post, though.

With that said, I do want to express that I think it's wonderful that this project is reaching direct descendants of systems from the 1960s, as illustrated by the aforementioned PR and this discussion.

[davepl]

I, for one, would LOVE to see results on big iron! If it can't run the entire docker suite, even a manual recompile of the C++ and Java versions would be cool just to get the numbers. If you do the primes_par.cpp solution I can put it in context of comparing it to several other hardware platforms.

[rbergen]

If it is possible to run a few solutions on z/OS (again), then I think it would be great to also include the COBOL implementation included in #333. That language is of course period correct for the early ancestors of the mainframes that now run z/OS.

[andrew890]

The numbers on z/OS are difficult to compare to other systems (even other z/OS systems) due to variability. Due to the way software pricing has evolved, z/OS system capacity is VERY granular. The difference between the fastest CPU i.e. single thread speed and slowest is around 20x. Then you can have from 1-190 cores.

On my system (single core VM guest) I get around 4800 passes. The performance relative to other platforms seemed to improve for larger numbers e.g. 1 billion, which might be due to larger CPU cache. I can't do any meaningful parallel testing due to the single core.

The main interest for me is to compare different languages on z/OS. I don't know COBOL, but #333 looks easy enough to run so I can do that and compare to C++ and Java.

I am planning to write up my results and I would like to make the code available for other z/OS people to try. I can do that in my own repository with a link back to the original, or contribute it here.

Maybe a show and tell directory could be created in the original repository for things that don't fit the main project? Another candidate would be the C solution that does the calculation at compile time - it's very interesting, but not proper drag racing to start at the finish line!

[rbergen]

I am planning to write up my results and I would like to make the code available for other z/OS people to try. I can do that in my own repository with a link back to the original, or contribute it here.

I think creating your own repository and sharing it in a "show & tell" post (that is an actual Discussions category) here would be awesome.

Maybe a show and tell directory could be created in the original repository for things that don't fit the main project? Another candidate would be the C solution that does the calculation at compile time - it's very interesting, but not proper drag racing to start at the finish line!

This repository's contents must target the amd64 and/or arm64 architectures (preferably both), because that is a requirement for inclusion in the benchmark runs. Also, at this time we can only accept new submissions if they either add a new language, improve the performance of an existing solution, or demonstrate the performance impact of a specific implementation approach (usually leading to a change in one of the key characteristics compared to other entries).
The C++ entry you are referring to is one of the latter kind. Due the fact it does "start at the finish line" as you call it, it is not an "official" entry for the actual drag race (it's marked as "unfaithful" for that reason).

As I said, I would love to have a Discussions post here that shares your work and results. I, for one, can't wait to see what they are.