Egghead models for soccer.
In the Dutch Correspondent I found an interesting article from 2019 about data and soccer. It seems that a scientific approach to managing a soccer team is more promising than a hiring a charismatic coach.
The scientist Sumpter at work
I am currently getting myself up to speed with AWS through an AWS Certified Solution Architect – Associate training.
As an architect with quite a bit of background in infrastructures, I was surprised by the level of minute infrastructure details that are still needed for the design of a simple virtualized environment. Even more so by and the amount of subsequent manual configuration to setup an environment. The level is no less than what is needed for a physical environment. My ignorance for sure, but I had expected the IaaS provider to hide much more detail from the user. What you see however is that server and network details up to things like CIDR blocks is still needed.
I guess there is an opportunity to create more high level infrastructure boilerplates. Around 2000 I was involved in many projects needing similar detailed design for the e-business infrastructures clients needed at that time. After having done that a few times we find that these infrastructure are in fact very similar. For a service like AWS I can similarly envision a boilerplate that needs only that entering of a limited number of functional and non-functional requirements for an setup like a web application infrastructure with a user registry, a scaleable set of application servers and highly available relational database management system.
(I am assuming IBM, Google, Microsoft and others are not different in this respect, but admittedly I have not checked all of them)
To be fair, AWS does have include a quite extensive documentation system available with reference architectures, technical guides and whitepapers. Here you can find blueprints for solutions, and guidance how to set up such solutions in AWS. Addin assets, tools, automations to support users in setting up such best practice configurations would probably be a great addition to AWS’ Architecture Center and/or Marketplace.
What’s your experience?
I listed my wishes for the mainframe. Can we have this please ?
For our software and hardware suppliers:
Continuous testing is not an option anymore. For the speed that modern agile software factories need, continuous functional, regression and performance testing are mandatory. But with mainframes, the cost of continuous testing quickly become inhibitive. Currently all MSUs are the same, and the new testing MSUs are driving the hardware and more importantly the MLC software costs through the roof.
Please not through yet another impractical License model (see later).
For reliable and manageable regression and performance testing, multiple configurations of test environments are needed, on real Z hardware. The emulation software zPDT or zD&T is not fit-for-purpose and not a manageable and maintainable solution for these needs.
Customers do not want to notice that development teams of IBM / CA/Broadcom / Compuware / BMC / HCL / Rocket / … are geographically dispersed (pun). Please let all your software developers in your company work the same way, on shared problem.
Everything sysplex-enabled by default please. Meaning, ready for data sharing, queue sharing, file sharing, VIPA, etcetera.
I do not mean Saas. I mean everything is scripted and ready to be automated. Everything can be engineered and parameterised once and rerun many times.
Want to boost innovation on the mainframe? Let’s have a publicly accessible mainframe for individual practitioners. And I mean for z/OS!
Give us more open source for z/OS. And the opportunity for the broad public to contribute. We need a community mainframe for z/OS open source initiatives.
Extend what z/OSMF has given us: APIs on everything you create. Automatically.
Yes there are more characters than those in codepage 037,and they are really used on mainframes outside the US.
(everything!)
Ok make it half an hour. Still too challenging? Ok, half a day. (Hide SMP/E for customers?)
There is a lot of things that are not useful nowadays anymore. For example, ISPF is full of it. For example Primary ISPF screen ISR@PRIM Option 1, 4, 5 7 can go. Many other things (print and punch utilities, really).
Remove customization options. Work on a standardized z/OS image. We don’t need the option to rename SYS1.LINKLIB to OURCO.CUSTOM.LINKLIB. Define a standard. If customers want to deviate, it is their problem, not all ours.
My goodness everyone has invented something to get code from the installation to their LPARs because there’s nothing there. Develop/define a standard. (Oh, and we do not need SMP/E on production systems).
For example stop (near) eternal downward compatibility (announce everything must be 64 bit from 2025 forward). Abandon assembler. Force customers to clean up their stuff too.
Ok if it applies to innovative/renovated applications. (But keep pricing SIMPLE please. No 25 new license models, just 1.)
Slower is not always worse, even into day’s agile world… Fast is good enough.
We need the next-gen CBT Tape.
Versus “this is how we have done things the past 30 years so it must be good”. Yawn.
Remove need for (unnecessary) IPLs, rebinds, restarts (unless in rippling clusters), …Kill all Exits (see later).
Remove assembler interfaces (anything is better than assembler). Replace with open APIs. Remove all (assembler) exits (see later).
By default.
As much memory as we can get (at a reasonable price). (Getting there, kudo’s.)
Smaller z/OS sites crave zIIPs to run the innovative tools like z/OSMF, ZOWE, Ansible, python, etcetera seamlessly.
Give us parameters, inifiles yaml, JSON, properties, anything other than exits. Better even: give us no options for customization. Standardize.
For our mainframe users:
Kill old technology. Assembler, custom build ISPF interfaces, CLISTs, unsupported compilers, applications using VSAM or worse BDAM, …
Modernise your DevOps pipeline. Use modern tools with modern interfaces and integrations. They are available.
Re-architect applications. Break silo’s into (micro) services while rebuilding applications. (Use Microservices where useful, not as a standard architecture)
Retire old applications. Either revamp or retire. If you have not touched your application for more than 2 years, it has become an operational risk.
Hire young people, teach them new tools, forbid them to use old tools. Yes I know we need 3270, but not to edit code.
Task your young people with technology modernization.
What am I missing?
Everyone thinks their own area of technical expertise is the most important.
The Data Architect thinks software solutions must be data-driven.
The integration architect thinks everything must be event-driven, or every interface must be a REST API.
The service management guys think that the CMDB is the center of the universe.
The cloud architect (if such a thing exists) thinks everything must be deployed in the cloud because the cloud is heaven.
We all forget that successful architectures are based on best practices. Quite universal best practices. Don’t tie everything tightly together (layering, loose coupling), do not make things complex (simplicity), etcetera. Technologies are not a goal. They are just a means. At best. Technologies are a necessary evil. You want as little as possible of it.
People around me are talking about using GraphQL, where they position this next to or opposed to REST API’s. I was not sure how these compare so I needed to find out.
In short, from 10000 feet, GraphQL is an alternatief for REST API’s for application programming interfaces. GraphQL provides more flexibility from several perspectives. Read more about that in the link below.
However, it requires a specific GraphQL server side infrastructure. This is probably also going to be a problem for large scale adoption. You can build a GraphQL client in a number of programming languages, but to serve as a GraphQL provider you need one of the few server side implementations.
So, a big benefit of REST API’s is that is is an implementation-independent interface specification that is easy to implement on your server side middleware. GraphQL would require your middleware to integrate one of the GraphQL implementations, or build one natively. This could be a matter of time and adoption, but currently I do not see a broad adoption.
Apollo, graphql-ruby, Juniper, gqlgen, and Lacinia are GraphQL implementations.
I found the article that best describes what GraphQL is on the AWS blog Comparing API Design Architectures – AWS (amazon.com). (I am not an AWS afficionado, it is just that the article best addressed what I wanted to know.)
This week I entertained a little talk about provisioning automation for z/OS. IBM has created a provisioning tool for z/OS that is part of the z/OS base. I talked about our experiences with the tool. It is changing to Ansible technology now. Next technology hop. Let’s talk tech again to refrain from doing things.
Later that same day I started a course called Google Cloud Platform Essentials.
Auch!!
We are somewhat behind on z/OS. That is a major understatement.
We done do tapes anymore, and MVSGENs, but it still feels like an upgrade from a horse to a steam engine.
Yet, I believe if done well the z/OS tools available today should allow us to catch up quickly.
There’s no technical obstacle. It’s only mindset.
Technical debt is a well-understood and well-ignored reality.
We love to build new stuff, with new technologies. Which we are sure will soon replace the old stuff.
We borrow time by writing quick (and dirty) code, building up debt.
Eventually we have to pay back — with interest. There’s a high interest on quick and dirty code.
Making changes to the code becomes more and more cumbersome. Then business change becomes more painstakingly hard.
That is why continuous renovation is a business concern.
Organisations run into trouble when they continue ignoring technical debt, and keep building new stuff while neglecting the old stuff.
Techies like new stuff, but if they are professional they also warn you for the old stuff still out there. You see them often getting frustrated with overly pragmatic business or project management, pushing away the renovation needs.
Continuous renovation must be part of an IT organisation’s Continous Delivery and application lifecycle management practice.
Making renovation a priority requires courage. Renovation is unsexy. It requires a perspective that extends the short-term horizon.
But the alternative is a monstruous project every so many years to free an organisation from the concrete shoes of unmaintainable applications. At best. If you can afford such a project. Many organization do not survive the neglect of technical debt.
The 80/20 principle also known as the Pareto principle applies to many areas.
The most common application of the principle is in the assessment of a project effort: 20% of the effort produces 80% of the job to be do done. Or the other way around, the last 20% of the work to be done will take 80% of the effort.
In IT, the principle applies also similarly to requirements versus functionality: 20% of the requirements determine 80 procent of the architecture. 20% of the requirements are the important ones for the core construction of a solution.
The principe thus tells you to focus on the 20% important requirements that determine the architecture. It helps you shrink the options you need to consider and prioritize and focus on the important parts of the solution.
The question is of course: which of the requirements are the important ones. The experience of the architect helps here. But in general you will realize while analysing requirements, if a requirement will need a significant change or addition to the solution.
A good book about the 80/20 principle is the book with the same name: The 80/20 Principle, by Richard Koch.
An example from an practitioner architecting an airline reservation system.
“The first time I (unconsciously) applied the 80/20 rule was in my early days as an architect. I was working with a team of architects on application infrastructure for completely new web applications. A wide variety of applications were planned to run on this infrastructure. However, it was not clear yet what the characteristics, the volumes, response time needs, concurrent users et cetera were for these applications, and that made it uncertain what we needed to cater for in this generic hosting solution.
So we decided to walk through the known use cases for the different applications.
We worked our way through four of the tens of applications. During the fourth we somehow could not come up with additional requirements for the application infrastructure. We realized that the rest of the set of applications would ‘only’ be a variety of one of the apps wealready looked at. So we had our 80% from looking at just 20%.”
This blog is now transitioning. When I started the blog I wanted to write about IBM mainframe technology, giving space to other readers, presenting a fresh view.
My intentions have changed, challenges have changed, and readers have changed.
After some posts expressing somewhat obverse standpoints of mine, readers reacted they wanted more of that. Also, in an earlier blog I shared snippets called ‘Principles of doing IT’, which got positive feedback. In this blog I will now bring these together. I will categorize my posts so the reader can easily filter what he wants so see. Yet, I give myself the freedom to keep posting in the order I like, and on the topic that I feel most urgently needs an obverse view.
I hope you enjoy. Please let me know what you think.
Niek
Snippet of shell script code to test if a directory exists in z/OS Unix System Services.
#!/bin/sh
DEST="/app/somedirectory"
if test -d "$DEST"
then
# It is a directory #
echo "Directory found ..."
else
# does not exist
echo "Directory does not exist"
fi