Pushing The Capacity Envelopes With IBM i 7.3
April 9, 2018 Timothy Prickett Morgan
Every piece of systems software has some sort of maximum capabilities reference, and these documents are interesting in their own right to help system administrators, programmers, and database administrators, as they case may be, figure out where they might hit a ceiling in terms of capacity or performance.
IBM has just published the Availability Maximum Capacities reference for IBM i Version 7 Release 3, which you can take a gander at here. As this reference correctly points out, coming close to the performance limits – or many of them at the same time – can cause outages. In this, we learn that up to 128 machines can be configured in a cluster, but which we presume that IBM means DB2 Multisystem, a database clustering technique that is over two decades old and that is akin to the Parallel Sysplex on its System z mainframes.
We also learn that the OptiConnect fibre optic links between cluster nodes can address up to 64 partitions, which is a neat limit, and that the TCP/IP stack, which IBM i inherits from AIX through a PASE runtime, can address around 100,000 configured objects and can have 65,535 ports and routes per system, and the alternative Systems Network Architecture (SNA) stack still lurks inside of IBM i and up to 25,300 peer-to-peer devices can be configured to an IBM i instance. I doubt very much anyone is even close to these theoretical limits. IBM i can have 15 system libraries and 250 user libraries, and no more than that, and there can be a maximum number of 1 million objects in a library and 2.15 billion objects across the root, QOpenSys, and user defined file systems in the Auxiliary Storage Pools (ASPs) number 1 through 32 on a single machine. The machine can, in theory, have 2.15 billion jobs in a subsystem, up to 32,767 of them can be active at the same time, and it can have no more than 35,600 disk arms and no more than 223 ASPs and a maximum load source size of 2 TB. No partition can have more than 64 TB of main memory, and no more than 192 processors with four threads each can be linked into a single NUMA image.
There’s a lot more in there, of course. These are just the ones I initially found interesting. It seems unlikely that many of these limits will change much in the coming years. But you never know.