By Akiko Hoshikawa
Every new generation of the IBM Z mainframe comes with excitement—not just for its technical advancements, but also for the unveiling of its sleek front panel design, often kept secret until launch. The latest member of the family, IBM z17, which was released in June, brings significant enhancements in AI capabilities and performance that directly benefit Db2 for z/OS workloads alongside that cool new front panel. Here’s what it means for you.
Smarter with AI: Accelerated Performance for Db2
The IBM z17 is powered by the new Telum II processor, which is specifically designed to handle AI workloads. The Telum II boasts:
- 8 cores running at 5.5 GHz, up from 5 GHz in z16
- A second-generation on-chip AI accelerator (zAIU) for intelligent, low-latency AI inferencing
Db2 leverages this AI acceleration through SQL Data Insights, allowing concurrent AI queries to run faster and more efficiently. With z17, each core can offload AI inference tasks to zAIUs—now even across chips within the same drawer. In our performance tests, concurrent AI semantic queries ran up to 24% faster on z17.
Bigger Cache, Better Efficiency
One of the biggest advantages of z17 is its 40% larger processor cache. For Db2 workloads, data movement dominates CPU cycles. This huge increase in cache provides significant benefits by moving more data closer to the CPU, which greatly minimizes processor cache misses.
In IBM benchmarks, z17 consistently outperformed z16 across varied Db2 workloads thanks to the combined benefits of larger processor caches and faster CPUs.
The Best I/O Is No I/O
IBM z17 supports up to 64 TB of memory per CEC, dramatically reducing dependence on I/O-intensive operations and boosting overall efficiency. Many features of Db2 for z/OS are designed to take full advantage of this memory capacity, including:
- In-memory buffer pools with PGSTEAL(NONE)
- Fast Traverse Blocks (FTBs) to avoid index traversals and get pages
- Index look-aside for more efficient index access
- Thread re-use with RELEASE(DEALLOCATE), High performance DBATs to avoid releasing locks and thread information across commits
- In-memory sort to avoid workfile processing
- Insert Algorithm 2 for efficient high-volume insert space search
- ...and more
These capabilities help Db2 to optimize processing by caching data or control structures in-memory to lower CPU consumption, reducing I/Os, and unlocking the full potential of z17’s large memory capacity.
zHyperLink 2.0: Breaking the I/O Bottleneck
zHyperLink is a short-distance, high-speed connection between IBM Z and DS8K storage. Db2 supports it for:
- Synchronous read I/Os from tables and indexes
- Synchronous write I/Os to active logs
Combined with IBM z17 PCIe Gen4 and the 10th-generation of DS8000 (DS8000 G10), zHyperLink (zHL) gains a significant performance boost. With the DS8000 G10, zHL now supports Db2 synchronous read I/Os for pages larger than 4KB, further increasing its value.
For active log writes, we’ve observed the following performance improvement using SAP Banking Service Account Settlement workload:
- Up to 40% faster log write throughput, which results in up to a 29% reduction in batch elapsed time when comparing zHL with G10 (DS8A50) against zHL with DS8950
- An additional 10% latency improvement with IBM z17 enhancements
For more details, see IBM Performance Report: zHyperLink for Db2 Logging with SAP.
What’s Next: AI and Observability on the Horizon
To further infuse AI into mainframe workloads and improve observability, IBM has announced the following future capabilities:
Spyre AI Accelerator
IBM plans to introduce Spyre, a PCIe-attached AI processor to bring generative AI to the mainframe. Spyre will enable on-premises deployment of watsonx Code Assistant for Z and watsonx Assistant for Z.
Evolving SQL Data Insights
Db2 13 introduced SQL Data Insights to uncover relationships in structured data. With z17’s integrated AI acceleration, IBM plans to expand this capability to process both structured and unstructured data together, which can increase the potential for uncovering new AI insights from more of your data.
Sensitive Data Tagging
IBM plans to deliver new AI capabilities to automatically discover and classify sensitive data in z/OS data sets, including Db2 data. These capabilities will leverage natural language processing on Telum II, helping enforce data security policies such as encryption.
Open Telemetry Support
IBM intends to support Open Telemetry(OTel) tracing across Db2, CICS, IMS, MQ, and IBM Data Server Driver for JDBC/SQLJ. This enhancement will provide end-to-end traceability across middleware products for faster issue resolution, even in complex hybrid environments.
Conclusion
IBM z17 is a tightly integrated platform that combines AI acceleration, a cache-rich architecture, and I/O innovations. Db2 for z/OS is fully optimized to leverage these advancements, delivering enhanced performance, deeper insights, and new possibilities for modern workloads.
Acknowledgment & Disclaimer:
Performance results are based on internal measurements conducted by the members of Db2 for z/OS Performance and IBM Z System Performance team under the controlled environments. The actual throughput or performance differences will vary depending upon many factors and no assurance can be given that an individual user will achieve results similar to those stated here.