In A Parallel Universe, There Is An IBM i Data Warehouse, Too
November 8, 2023 Timothy Prickett Morgan
Over the many years that we have been watching the Power Systems business – which is all of them, starting with the original RS/6000 announcement in February 1990, the first announcement we ever attended and at IBM’s original headquarters at 590 Madison in New York City – we have seen time and time again how Big Blue has rolled out parallel cluster data warehouse configurations for its AIX customers.
Somehow, the IBM i platform and predecessor OS/400 platform have never been shown the same respect. Despite the fact that there are midrange and large enterprise customers who would very likely enjoy building a data warehouse using the same, identical technology that they use in their production online transaction processing systems. Instead, OS/400 and IBM i customers have had to try other things to get data warehouse functionality, like move subsets of their data to X86 clusters running Windows Server and SQL Server from Microsoft. And in case you are getting the idea that we are pretty annoyed by this, let us assure you that we most certainly are.
We were poked by this lack of a native data warehousing platform from Big Blue for OS/400 and IBM i by announcement letter AD23-0855, which came out this week and which covers the launch of what is called IBM Power Private Cloud Rack for Db2 Warehouse. This is a new Power Systems parallel database cluster based on Power10 servers and the Unix-Linux-Windows variant of the Db2 database with data warehouse extensions. IBM says in its spec sheet for the database cluster that it is a follow-on to the PureData System for Operational Analytics (PDOA) and the IBM Integrated Analytics System (IIAS), and specifically it is based on relatively inexpensive and yet powerful Power S1022 nodes running the Red Hat OpenShift implementation of the Kubernetes container environment atop Linux. Presumably both the management nodes and the database nodes are running Linux and not AIX; we know they are not running Windows Server or IBM i, of course.
The new Power10 database cluster scales from a third of a rack with three worker nodes to eight racks running 77 worker nodes, scaling from 48 to 1,656 compute cores and from 2.59 PB of user data space on the compute nodes, like this:
There are additional storage nodes in the Db2 Warehouse cluster based on Power10 that add another 4.12 PB of capacity for additional capacity.
Remember: This is all based on that very affordable and very modest Power S1022 server, which we detailed back here in August 2022. The Power S1022 has two sockets, and each socket has a pair of Power10 CPUs sharing that socket to boost the overall core count and I/O and memory capacity of the system. It packs a lot of compute, memory, and storage into a 2U form factor.
And we think that because the Power S1022 can run IBM i 7.5 and it can make use of the Db2 Multisystem parallel clustering features of OS/400 and IBM i, that Big Blue should create a version of the Db2 Warehouse that is based on the IBM i and Db2 for i stack instead of the Linux and Db2 for Linux stack. Why not?
According to the Db2 Multisystem documentation, which you can see here, such a parallel warehouse implementation based on the IBM i stack would only scale to 32 systems, which is a little less than half of the scale of the Linux version that IBM just announced. Somewhere around three and a half racks of iron in the chart above and somewhere around 1.1 PB of user data capacity in the warehouse, with 682 cores and 5,456 threads ripping across a cluster that could, using partitioned database tables spread across those 32 nodes, run queries 32X faster than if they were run on one Power S1022.
Think of all of the Microsoft SQL Server or Oracle Database warehouses that this might replace. Think of the incremental revenue IBM could make and the incremental revenue that IBM i shops might be able to drive in their own businesses with a data warehouse that was built on the same technology they already use and know well.