Let’s Try Converged Power Infrastructure One More Time
April 8, 2019 Timothy Prickett Morgan
Do you remember the Flex System modular servers launched seven years ago this month? These were the innovative machines that Big Blue sold off to Lenovo about two and a half years after they were launched and they were ramping? Do you remember the PurePower follow-ons to these that came out in May 2015? Or did we all just imagine that happened?
These modular machines, which were somewhere halfway between a rack server and a blade server, were put into preconfigured stacks and as the PureFlex system had cloud automation software to create a private cloud and then had PureApplication variants that took it one step further, creating scripting tools DevOps-style, called patterns, to automate the provisioning of virtual compute, storage, and networking for and the automagic deployment of applications. After the System x division was sold to Lenovo in early 2015 and IBM wanted to focus on its Power-based iron, it created a similar stack of hardware and software – what is variously called converged systems or integrated systems – called PurePower and based on Power S822 and Power S822L nodes using (as the names suggest) Power9 processors. In October of that year, IBM i was added as a supported operating system on the Power S822 nodes in the PurePower clusters, allowing for a private cloud to provision IBM i, AIX, and Linux on the same hardware substrate.
An announcement from IBM last week got me to thinking about all PurePower and predecessor effort to create a converged platform to peddle into the heterogeneous datacenters of the world. This particular announcement was not of huge consequence. In announcement letter 219-217, IBM promised that it would be updating the PureApplication W3500 and W3550 appliances, which are Xeon-based clusters to run the full-on PureApplication stack. IBM also said that it would allow the PureApplication automated systems to be able to deploy Red Hat OpenShift, its Kubernetes environment for Docker containers, as a pattern, and ditto for IBM’s homegrown Cloud Automation Manager Kubernetes/Docker stack.
The PureApplication W3700 appliances, which were based on Power7+ iron, have not been upgraded to Power8 or Power9 servers as far as we know, and presumably the PurePower cluster based on plain vanilla Power Systems rack servers has replaced them. As far as we can tell from IBM’s sales manual and announcement letters, these PurePower clusters have not been updated to Power9 iron, either. But with Red Hat being acquired by IBM for $34 billion, we could see a kind of stackable, rackable Power-Linux superstructure offering in the not too distant future, and one we hope that treats IBM i as a peer to Linux.
We spend a lot of time thinking about how to make the Power Systems business stronger, and there are two approaches and both are reasonable. One is to take a single Power server and run IBM i on the core partitions and then bring in Linux infrastructure workloads that replace Windows Server functions in the datacenter. The other is to create a hybrid rack that includes Power nodes and X86 nodes – and Arm nodes if there is a call for them – and put them all under the same management domain and literally run IBM i, AIX, Linux, and Windows Server on the machines that are best suited to the task. IBM tried this with BladeCenter blade servers starting in 2001 and again with the Flex System iron again in 2012. There were some successes in hybrid systems, and certainly some capital expense and operational expense saving on these converged platforms. But it is not like they took the market by storm.
The acquisition of Red Hat gives IBM a very good chance to start replacing Windows Server and bringing all kinds of applications onto Linux running natively on Power.
Certainly all of the major open source projects relating to data analytics and machine learning frameworks in the world are designed for Linux, and IBM can do a better job leveraging this by expanding the Red Hat portfolio to include all kinds of things, including Hadoop, Spark, Redis, Cassandra, and a litany of software all supported by IBM through the expanded Red Hat organization. Every important database and datastore – MySQL and MariaDB for straight up SQL serving; TimescaleDB and InfluxDB for time series databases; MongoDB and Couchbase for document databases; Tigergraph, GraphDB, and Neo4j for graph databases; and funky stuff like CockroachDB (inspired by Google Spanner) and FaunaDB (inspired by Twitter) that are distributed databases that are made to scale across multiple geographies. The point is, you can support Oracle. Db2, and Db2 for i for traditional transaction processing, but you need all this other stuff to help the Power Systems business grow. Anything that makes Power Systems stronger makes IBM i last longer, and these should all be supported atop RHEL and, I think, turned into distributions that are supported by the combined Red Hat/IBM organization.
IBM needs to be thinking on a much larger scale than Linux operating systems, JBoss middleware, and Kubernetes/Docker container infrastructure. The beauty is that all of these projects are open source, and any of them can offer an alternative provider of tech support. Or have their teams acquired to build out the software portfolio. Why not? IBM is not going to be able to sell mainframe middleware forever. IBM i might last forever, but it cannot alone support the investments required to create a Power processor and server platform. But the revenue streams from all of these kinds of projects, with the Big Blue and Red Hat dynamic duo, could make IBM formidable in open systems on OpenPower.
Stranger things have happened. Like IBM buying Red Hat, for instance.