I recently attended an IBM update on its analytics business.
One of the core themes was simplicity. As we all know enterprise products generally, and IBM products specifically, are not known for their ease of use. So when I heard that IBM middleware had a new approach – dubbed Download and Go – so a customer can install a product and “get to value” in 15 minutes I wasn’t too excited. Installer technology… and frankly, we’ve heard it all before. Of course IBM needs to get better with free to download products that are easier to install, if it wants to widen the digital funnel, and get Purchasing out of the way of developers and engineers.
RedMonk being RedMonk my first question was why not “upload and go”? If you’re talking about analytics then cloud is the context that really matters. Why isn’t IBM focusing on making it easier for customers to get their data into the cloud for analytics and AI/ML? In conversations during our demo breakout however things took a turn for the interesting.
Turns out IBM is going to Dockerise its software products and then make packages available which include all the open source and proprietary componentry, with Kubernetes managing deployment to on prem clusters. These don’t have to be big clusters – one of the new packages which will be announced this Thursday is designed for 3 node deployment. That package will be on prem first and while I am skeptical about that, IBM claimed customers were most interested in on prem deployment first.The Kubernetes approach also means IBM has portability out of the box for x86, POWER (including zLinux) and Nvidia GPUs.
IBM’s Data Science Experience platform is already enabled with the new approach, as per the screenshot above. Hit the landing page and you can download the Docker app to your laptop, ready for deployment to a Kubernetes cluster or run it in the cloud. The package contains everything the developer needs for RStudio, Jupyter, Python, Scala, including Spark, with Tensor Flow in the pipeline.
In case you didn’t get the memo IBM’s Bluemix platform is also now increasingly oriented to Kubernetes rather than CloudFoundry. IBM will support CloudFoundry for existing customers, but will be shifting future investments to Kubernetes, and technologies such as OpenWhisk and Istio. IBM’s Cloud platform technologists are very very happy with Kubernetes performance, particularly in terms of resilience, so far.
One reason I found the approach interesting is that I have spoken to a couple of enterprise IT organisations recently that are in a position of managed decline with respect to their own server real estate, as workloads move to the public cloud, but actually have a lot of servers which they’ve already paid for, looking for workloads. Enterprises have to do something with that investment. Analytics is one of those workloads where server huggers could find a bit of a sweet spot.
I recently wrote about that giant sucking sound at Hadoop moves into the cloud. One of the reasons Hadoop workloads are moving into the cloud is the sheer administrative hassle of managing clusters and software stacks. IBM’s recent announcement of a deal with Hortonworks shows that customers don’t want to deal with hassle and incomplete software stacks. Indeed they want simplicity.
The Kubernetes decision for data oriented workloads around technologies such as Spark is not particularly great news for Mesosphere, which has pitched itself as the container/cluster scheduler for big data, with partners including Lightbend.
“Elastically run containers and data services at scale, with complete hybrid cloud portability.”
It will be interesting to see how far IBM goes with the Kubernetes model. Could it even retrofit older versions of its middleware products, allowing for example, rolling upgrades across Kubernetes clusters, and refactoring before database version upgrades?
In summary, this is an aggressive application of Kubernetes for on premises workloads. IBM tried OpenStack as the basis for a hybrid cloud, then CloudFoundry, but seems to have settled on Docker/Kubernetes as the stack that will work best as a hybrid. Like Red Hat with OpenShift IBM took a few cycles to get it right, but seems to have settled on an approach that makes sense for enterprise customers. Talking of Red Hat, another potential implication of the Docker/Kubernetes approach is a reduced reliance on specific Linux distros.
I will retrofit this post on Thursday with some details of the specific stacks being deployed.
[update…. so the news is now official.
When I posted this story last week, Steve Citron Pousty was a sceptical.
so @kubernetesio. On Prem. Less Distro. Whatever Next? Download and Go http://redmonk.com/jgovernor/2017/06/20/kubernetes-for-on-prem-simplicity-download-and-go/ …
Steve’s point is well made, but Db2 is different than the technologies above – it is certainly not cloud native, for one thing. Nor is it open source. Db2 Developer Community Edition is now available with an installer based on Kubernetes (though the docs just mention Docker, so i am going to be looking at these in more detail). I would also argue the consumption models here are different. OpenShift is a PaaS, with a menu of options. Db2 DCE on other hand, is designed to be installed in standalone fashion. JSON support and free-download-for-developers are not going to change the game regarding Db2 adoption – IBM has offered free versions of the database before. The container-based approach to delivering IBM middleware products, making them more convenient for on prem deployment is an interesting move. This is not a Bluemix on prem model, but a separate IBM business unit also betting on the standard for container orchestration. IBM isn’t alone. Oracle Database 12c Enterprise Edition is now available in the Docker store, for example, for free to developers. SAP Labs is also looking at Kubernetes for on prem deployment.
While containers are definitely not a one for one replacement for virtualisation or OS deployment, the strong interest of the enterprise leviathans has implications for both VMware and Red Hat Linux. The container simplicity story is winning, and Kubernetes is shaping up as a major technology for hybrid cloud.
IBM, Oracle and Red Hat are clients.
(Read this and other great posts @ RedMonk)