Porting Go to Linux on IBM Z Architecture | Hacker News Reader

Porting Go to Linux on IBM Z Architecture

groups.google.com

66 pointsdiakritikal24 comments

IBM has been recently promoting Docker on z. IBM argues the high I/O capabilities of their platform are a good fit for the container model. (I don't disagree with them. The many-application-on-big-hardware model is not a new concept to IBM. Hell, they invented it.)

Anyways... Since Docker is written in Go, the only way Docker has been available for s390x was by building with gccgo. My guess is that native Go support is in the interest of Docker on z and would be a worthwhile investment for IBM.

IBM is also the second-biggest contributor to Cloud Foundry after Pivotal (for whom, disclaimer, I work). Almost all new CF code is written in Go.

I expect they're going to position mainframes as an ideal platform for CF installations (under a title like "BlueMix Z").

> Docker on z.

Docker on Linux on VM on LPARs on z.

Nobody say it...

More likely Docker on Linux on LPAR. At that point you'd struggle to see the value of VM, other than making some of the device management/maintenance easier.

Pascal once ran on IBM nainframes for an easy to learn, read, and compile solution. Go is latest in Wirth philosophy of languages. Good it's going on mainframe, too.

Even better is that it's often used for I/O intensive (eg networking) applications. Mainframes' Channel I/O blows traditional server architecture away in throughput. Should give a boost.

Go is a Wirth language? Do you have a source for that?

I said "Wirth philosophy." It's his style of a language in action but invented by others. Greisemer had enjoyed the simplicity and productivity of Wirth's Oberon language at a point in the past. He wanted to re-create and expand on that in their C++ replacement. Source is below:

https://sourcegraph.com/blog/live/gophercon2015/123645585015

Yes, I got that. Thanks for the link!

> IBM Linux on Z Open Source Ecosystem Center of Competency

Man, what a name for a group!

LOL. My observation as well. Truly a name that IBM would invent.

They're in a pissing match with DOD on this kind of thing, apparently...

I enjoy SAP Center of Excellence

> Going live with SAP enterprise applications is only the end of the beginning. In order to thrive after go-live, you need the right organization to drive continuous business improvement.

http://michaeldoane.com/SAP_Center_of_Excellence.html

They also created a native OCaml backend for IBM Z very recently: https://github.com/ocaml/ocaml/pull/275

What is Z exactly if I may naively ask? I've heard it's name but I'm not familiar.

CPU architecture of contemporary IBM mainframes. The 64-bit successor to the 24-bit/31-bit IBM System 360/370/390.

IBM is really trying to make the mainframe "cool"

since when is the mainframe not cool?

hip or trendy would have been better words (instead of cool) in my original post I think

Won't happen and thank god too. The hip and trendy crowd have no idea how to do long term support. JavaEE is the farthest from "hip and trendy" but you can fire up a 15 year old WebSphere application on brand new hardware and it'll work better than the day it was written, with full first-party support for every component that was supported from day one. I've seen production code REXX and JCL code from the 1970s which ran System/360 hardware with actual decades of uptime since the IBM mainframe was 100% redundant (literally every component from the DASD to the RAM and CPUs, which all supported hot-swappability for upgrades and failover, all within a 42U). It boggles my mind at how robustly these systems were designed. I.e., the original 360 code was based on a weird-ass 31-bit architecture. Now z/10's can run in 64 bit mode, 32 bit, 31 bit[1], or 16 bits with full support for every single application written from (theoretically 1964, but I've only seen code from the 70s still running) all the way up to Clojure and Scala if you want it.

Take a look at all those hip Ruby frameworks and look at when Yehuda Katz last made a commit to Merb and you'll figure out why CIOs still buy SAP and pay for Oracle. IBM is lucky not to deal with the 'hip and trendy' demographic- it'd just further damage their already faltering image in the corporate world.

The real market share they should be trying to pickup is not the frivolous Web 3.0ish junk. They instead should be pushing out a POWER8 + DB2 + fault-tolerant (i.e., milliseconds failover, not seconds) for under 250k with an ERP attached to it (partner with SAP Business One, 98% of the businesses doing under 250MM in gross per annum don't need more htan that). AIX 7 already has "Cluster-aware" software that does this, they just need to include it by default for up to 4U. THEN, start price gouging. Market it under the name "NEVER/down", partner with PureStorage or one of the new "EMC-killers", and aggressively market it like they did during the Watson hype (tennis ad-buys, advertising at LeMons, etc).

[1] https://en.wikipedia.org/wiki/IBM_ESA/390

I often see odes to ibm mainframe reliability. I'm piqued. Is playing with POWER or mainframe stuff edifying for a regular person, and if so, can a regular person obtain them?

I had some access to a mainframe this year and learned the basics of z/OS. Things do feel very restricted and slowish, you really notice that it comes from a batch background, and interactivity came later. A UNIX user might find it a bit frustrating. They can be interesting and do have their charms though. Interesting for you, on the real mainframe, due to the restrictions etc, there was only a limited number of things I could really do on it. Just the stuff I needed for my job and not much more. I actually learned more from reading "the book". I suggest just reading that and then you will have a good idea about whether you want to pursue access on the real machine or not. Free to download: http://www.redbooks.ibm.com/abstracts/sg246366.html

> you can fire up a 15 year old WebSphere application on brand new hardware and it'll work better than the day it was written, with full first-party support for every component that was supported from day one.

Spoken like someone who has never tried picking up code that worked fine on WAS 6.1 and tried running it on 8.5.

Actually, I had to do that very recently. The difference between IBM and Ruby was I called in with my client's license information and within 20 minutes my IBM id had access to a WAS 6.x download. My client upgraded all their hardware (this was from old iSeries AIX to a new POWER setup) and had zero-downtime after completing the transition. The only difference was a significant increase in speed. WAS 6 still is being actively supported and hot-patched for security which is what my clients want.

You're right though - moving up major releases can break things sometimes. I spent another couple of months forward-porting to Liberty so they could run 8.x but honestly most of the breaks I dealt with was dealing with third-party functionality that the old off-shore team decided to throw in. Regardless, during that period of time, my clients business was fully operational on that new hardware with a secure, seamless transition.

I guess you sort of get the same 'future proofing' these days with VMware (the analog is basically: having old hardware, then buying new hardware and vMotioning your production instance to the new VMware ESX cluster), but you certainly didn't have that option 15 years ago, when my client was choosing what environment to run his oil/commodities firm. And that's just support on the hardware stack, rather than the 10+ year support that most businesses want as a guarantee.

My first gig out of school was a cobol job, there were actually a handful of 100k+ LOC programs written in 1969 running, this was '03.

Reply on news.ycombinator.com